Merge pull request 'refactor(client): вставка markdown через канонический пакет + удаление md-слоя editor-ext (#347)' (#498) from refactor/347-client-md-paste into develop

Reviewed-on: #498

.env.example

@@ -124,6 +124,40 @@ MCP_DOCMOST_PASSWORD=
 # MCP_TOKEN=
 # MCP_SESSION_IDLE_MS=1800000
 #
+# --- MCP collaboration write path: concurrency + rights-staleness (#449) ------
+# MCP content writes (update_page, insert/replace nodes, comments-in-body, etc.)
+# go over the collaboration websocket and are serialized PER PAGE by an
+# in-process mutex (a module-level Map, one promise-chain per page UUID). This
+# guarantees no two MCP writes on the SAME page overlap and clobber each other.
+#
+# DEPLOY REQUIREMENT — SINGLE INSTANCE or STICKY SESSIONS. The mutex is
+# process-local. Behind a multi-replica load balancer WITHOUT sticky sessions,
+# two replicas can each "hold" the lock for the same page at the same time and
+# serialization is silently lost (concurrent full-document writes race on the
+# live Yjs fragment). Run the MCP/app as a SINGLE instance, OR pin a page's
+# traffic to one replica (sticky sessions / consistent hashing on page id). The
+# same constraint applies to the RAM-only stash_page blob store above. There is
+# deliberately no cross-process (e.g. Postgres advisory) lock yet — this is a
+# CONSCIOUS documented constraint, not an oversight (#449).
+#
+# To reduce connect-storms the write path caches ONE live collab session per
+# (wsUrl, page, token). Tunables (all optional; defaults are safe):
+# MCP_COLLAB_SESSION_IDLE_MS=60000     # idle TTL, reset per op; 0 disables cache
+# MCP_COLLAB_SESSION_MAX_ENTRIES=32    # LRU cap on cached sessions
+# MCP_COLLAB_TOKEN_TTL_MS=300000       # per-client collab-token cache (5 min)
+#
+# RIGHTS-STALENESS TRADE-OFF. A cached collab session writes under the token
+# captured at CONNECT time, and the collab-token cache reuses a token for its TTL.
+# So if a user's access to a page is REVOKED, MCP writes on an already-open
+# session may keep succeeding until the session ages out. MCP_COLLAB_SESSION_MAX_AGE_MS
+# is the HARD lifetime (checked at each acquire) that BOUNDS this window: after it,
+# the session is torn down and the next write re-auths with a fresh token, picking
+# up the revocation. Default 10 min. LOWER it to shorten the revocation lag at the
+# cost of more reconnects; RAISE it to reduce reconnects at the cost of a longer
+# stale-rights window. There is intentionally no push-based cache invalidation on
+# a rights change — this bounded window is the accepted trade-off (#449).
+# MCP_COLLAB_SESSION_MAX_AGE_MS=600000
+#
 # BLOB SANDBOX (stash_page). An in-RAM, process-local store that hands large page
 # content + images to an external consumer WITHOUT bloating the model context or
 # requiring Docmost auth. The stash_page tool serializes a page, mirrors its

.github/workflows/develop.yml

@@ -157,6 +157,12 @@ jobs:
       - name: Build prosemirror-markdown
         run: pnpm --filter @docmost/prosemirror-markdown build
 
+      # docmost-client.loader.ts type-imports from @docmost/mcp (issue #446); its
+      # build/ is gitignored and `test:e2e` type-checks, so build it here or tsc
+      # fails with TS2307 (mirrors the e2e-mcp / mcp-server-parity jobs).
+      - name: Build mcp
+        run: pnpm --filter @docmost/mcp build
+
       - name: Run migrations
         run: pnpm --filter ./apps/server migration:latest
 

AGENTS.md

@@ -5,6 +5,139 @@ repository. It has two layers: **how to run a task end-to-end** (the
 sections below), and **how the codebase is built** (the technical sections
 further down, formerly in `CLAUDE.md`).
 
+## ARCHITECTURAL INVARIANTS — NON-NEGOTIABLE
+
+THE TEN RULES BELOW ARE HARD CONSTRAINTS. Each one was paid for with a real
+production incident or a multi-PR bug chain in THIS repository (cited inline).
+They override convenience, deadlines and "it's just a small feature". A PR that
+violates any of them MUST be rejected in review regardless of how good the rest
+of it is. If a task genuinely seems to require breaking one — STOP and raise it
+with the owner; do not code around it.
+
+### 1. EVERY BUFFER, CACHE, HISTORY AND PAYLOAD HAS AN EXPLICIT SIZE BUDGET
+
+Nothing accumulates unboundedly. A row/item cap is NOT a byte cap. Anything
+replayed to a model, buffered in memory, persisted per step, or refetched by a
+poll must state its budget in bytes/tokens and enforce it. Rewriting a growing
+structure in full on every increment is FORBIDDEN — append or diff instead;
+O(n²) write/serialize patterns do not pass review.
+(Paid for by: full-row rewrite on every agent step — hundreds of MB of Postgres
+writes per 50-step run, with every tool output serialized twice; unbounded
+history replay killing long chats on the provider context window; 32 MB replay
+buffers per active run.)
+
+### 2. EVERYTHING LONG-RUNNING TERMINATES BY CONSTRUCTION
+
+Every run / row / session / lease / subscriber / queue entry must define AT
+DESIGN TIME: its owner; every terminal state; who writes the terminal state on
+EVERY path (success, error, abort, disconnect in each phase, process restart);
+retries for the terminal write; and a periodic sweeper that does not depend on
+a reboot. A best-effort terminal write with no retry and no sweep is FORBIDDEN.
+(Paid for by: assistant rows stuck 'streaming' forever; runs stuck 'running'
+409-locking their chat until a restart — the #183/#184 follow-up chain.)
+
+### 3. EVERY AWAIT IS CANCELLABLE AND DEADLINED; NEVER BLOCK THE EVENT LOOP
+
+Every async step inside a request or agent turn honors the turn's AbortSignal
+AND a wall-clock deadline — including in-app tools, lock queues and pagination
+loops, not just external calls. Synchronous CPU work beyond ~50 ms goes to a
+worker_thread. Promise.race DOES NOT cancel synchronous work — using it as a
+"timeout" for sync computation is forbidden (the timer only fires after the
+event loop is free again, i.e. after the damage is done).
+(Paid for by: in-app tools ignoring abortSignal and writing pages AFTER Stop;
+the synchronous ELK layout freezing every SSE stream in the process; the
+step-0 MCP handshake hang — #397.)
+
+### 4. ONE SOURCE OF TRUTH; EVERYTHING ELSE IS A REBUILDABLE CACHE
+
+Postgres is the authoritative state. Every in-memory structure (registries,
+caches, client stores) must be reconstructible from the DB and treated as
+lossy. The client renders SERVER-DECLARED state — "a run is active" is a server
+fact delivered as data, never inferred from side signals (204 vs 2xx, the
+flavor of a disconnect). A new feature must name the owner of each piece of
+state before implementation starts.
+(Paid for by: the strip/restore resume machinery, silently frozen UIs and
+ghost sends after unmount — the #381→#432→#456 chain.)
+
+### 5. STATE MACHINES ARE EXPLICIT — ONE-SHOT FLAGS ARE FORBIDDEN
+
+A complex lifecycle (chat thread, resume/reconnect, run) lives in a named-state
+automaton (reducer / enum) where every state has an owner and a rendered
+representation — including the failure states. Adding a boolean ref that one
+callback arms and another reads-and-clears is FORBIDDEN in the AI-chat client.
+New behavior = a new named state + explicit transitions, and the interruption
+matrix (disconnect in each phase × restart × stop × supersede) is enumerated at
+design time, not discovered one incident at a time.
+(Paid for by: 26 one-shot useRef flags in chat-thread.tsx and the drip of
+"one more missing transition" across #381→#386/#389→#432→#456.)
+
+### 6. NO NEW MODE FORKS; A FLAG IS FOR ROLLOUT, THEN IT DIES
+
+A behavior flag that forks a code path must ship with a written sunset
+condition; stacking a new flag onto the existing matrix without deleting or
+scheduling an old one is forbidden. While a temporary fork exists, BOTH sides
+must share identical lifecycle handling (abort semantics, error listeners,
+concurrency gates) — asymmetric forks are outlawed.
+(Paid for by: legacy vs autonomous divergence — the one-active-run gate and
+the socket 'error' listener each existing on only ONE side; 2^4 flag
+combinations each with different abort semantics.)
+
+### 7. NO HAND-SYNCED MIRRORS — CODEGEN OR A CI PARITY TEST, NOTHING LESS
+
+Two copies of the same knowledge (schema, tool registry, glyph map, probe
+body, hash/normalize algorithm, label list) require either generation from a
+single source or a CI test that FAILS on drift. A "mirror this change over
+there" comment is NOT a guard and does not pass review.
+(Paid for by: #293 — three drifting converter copies losing data; #447 —
+REGISTRY_STAMP covering only one of the mirrored files; ~10 still-unguarded
+mirrors across the MCP layer.)
+
+### 8. CACHES, HEADERS, BUFFERS AND FSM TRANSITIONS GET AN INTEGRATION TEST OF THE OBSERVABLE PROPERTY
+
+A unit test of a pure helper DOES NOT COUNT for these. Test the real header on
+the real HTTP response, the real cache hit under real token sources, the real
+transition under a really-killed socket. If the observable property cannot be
+tested, the design is wrong — fix the design, not the test.
+(Paid for by: #431→#439 — a cache keyed on a fresh-per-call JWT, so it NEVER
+hit and became prod incident #435 while its unit tests stayed green; and by
+the #352→#455 immutable-cache header silently overwritten by a framework
+default AFTER the unit-tested code ran.)
+
+### 9. CLIENT INPUT IS HOSTILE UNTIL VALIDATED — ALSO BEFORE PERSISTENCE
+
+Anything from the browser (message parts, ids, titles, selections, flags) is
+validated/sanitized BEFORE it is persisted into a row that will later be
+replayed into a prompt, a converter or another subsystem. A poisoned row must
+never be able to permanently brick a chat or a page on every subsequent read.
+(Paid for by: unvalidated UIMessage parts persisted verbatim — one bad row
+500s the chat on every later turn; #159 client-spoofed page titles; #388
+selection re-sanitized server-side for the same reason.)
+
+### 10. FAILURES ARE LOUD AND SPECIFIC; SILENT DEGRADATION IS FORBIDDEN
+
+Extends the error convention below: a fire-and-forget write is allowed ONLY
+with a metric or a greppable ERROR log; a degraded mode (dead cached MCP
+client, stopped poll, exhausted retries, evicted buffer) must be VISIBLE to
+the user or the operator. A feature that can quietly stop working — a frozen
+"streaming…" UI, a poll that silently gives up, a cache serving corpses — does
+not pass review.
+(Paid for by: the degraded poll's silent 10-minute death leaving a forever-
+"streaming" answer; dead MCP clients served from cache while every external
+tool call failed; #435 being caught in minutes ONLY because metrics — #403 —
+existed.)
+
+## Default skill for feature design
+
+For any feature-design request — the user hands over a raw feature idea, asks
+to design or think through a feature, or to draft an issue («спроектируй»,
+«продумай фичу», «составь ишью», "design X", "write an issue for X") — invoke
+the `orchestrator-feature-designer` skill (Skill tool) BEFORE any other work.
+It is the default operating mode for design work in this repository: research
+→ design checklist (R1–R10) → forks resolved with the human → adversarial
+self-attack → filed PR-sized issues. Do not design features or write issues
+ad-hoc while this skill is available. This does not apply to non-design work
+(bug fixes, reviews, retrospectives, refactors already specified by an issue).
+
 ## Task lifecycle
 
 ### 1. Start: sync with develop
@@ -201,7 +334,7 @@ pnpm workspace (`pnpm@10.4.0`) orchestrated by **Nx**. Four workspace packages:
 | `apps/client` | `client` | React 18 + Vite + Mantine 8 + TanStack Query + Jotai | SPA frontend |
 | `packages/editor-ext` | `@docmost/editor-ext` | Tiptap/ProseMirror | Shared Tiptap node/mark extensions, imported by both the client and the server |
 | `packages/mcp` | `@docmost/mcp` | MCP SDK, Tiptap, Yjs | Standalone MCP server, also bundled into the server at `/mcp`. Consumes the shared converter/schema from `@docmost/prosemirror-markdown` (#293) — it no longer carries its own vendored converter/schema copy |
-| `packages/prosemirror-markdown` | `@docmost/prosemirror-markdown` | Tiptap, marked, jsdom | The single, canonical ProseMirror↔Markdown converter + Docmost schema mirror (#293). Consumed by `mcp`, `git-sync`, AND `apps/server` (server-side markdown import/export, #345); there is exactly ONE copy of the converter now |
+| `packages/prosemirror-markdown` | `@docmost/prosemirror-markdown` | Tiptap, marked; jsdom (Node only) | The single, canonical ProseMirror↔Markdown converter + Docmost schema mirror (#293). Consumed by `mcp`, `git-sync`, `apps/server` (server-side markdown import/export, #345), AND `apps/client` (markdown paste/copy + AI-chat render, via the `browser` entry — native `DOMParser`, no jsdom in the client bundle, #347); there is exactly ONE copy of the converter now |
 
 `build` targets are Nx-cached and dependency-ordered (`dependsOn: ["^build"]`), so `editor-ext` builds before the apps. `nx.json` sets `affected.defaultBase: main`.
 
@@ -327,7 +460,7 @@ The API server is a Fastify app with a global `/api` prefix (`main.ts` excludes
 ### Client structure
 Vite SPA. Code is organized by feature under `apps/client/src/features/*` (mirrors the server domains: `page`, `space`, `comment`, `ai-chat`, `editor`, …). Conventions:
 - **TanStack Query** for server state (one `queries/` file per feature), **Jotai** atoms for local/shared UI state, **Mantine 8** + CSS modules (`*.module.css`) + `postcss-preset-mantine` for UI.
-- The editor is Tiptap; shared node/mark extensions live in `packages/editor-ext` and are imported by **both the client and the server** (collaboration, schema, `canonicalizeFootnotes`) — editor schema changes often need to be made in `editor-ext`, not just the client. Server-side markdown import/export no longer lives in `editor-ext`: it goes through the canonical converter (#345, see below). The ProseMirror↔Markdown converter and its Docmost schema mirror now live in a SINGLE package, `@docmost/prosemirror-markdown` (#293), consumed by `mcp`, `git-sync`, and `apps/server` (#345) — do NOT reintroduce a per-package copy. `editor-ext` is the upstream source of the Tiptap schema; the package's `docmost-schema.ts` mirrors it and a serializer-contract test (`packages/prosemirror-markdown/test/serializer-contract.test.ts`) guards the boundary (every schema node must have a converter case), so a drift surfaces as a failing test rather than silent divergence. For the converter's property-testing and counterexample→fixture process (P1–P4 invariants, the `PROPERTY_SEED`/`PROPERTY_NUM_RUNS` knobs, and the nightly fuzz workflow), see `packages/prosemirror-markdown/README.md`.
+- The editor is Tiptap; shared node/mark extensions live in `packages/editor-ext` and are imported by **both the client and the server** (collaboration, schema, `canonicalizeFootnotes`) — editor schema changes often need to be made in `editor-ext`, not just the client. Server-side markdown import/export no longer lives in `editor-ext`: it goes through the canonical converter (#345, see below). The ProseMirror↔Markdown converter and its Docmost schema mirror now live in a SINGLE package, `@docmost/prosemirror-markdown` (#293), consumed by `mcp`, `git-sync`, `apps/server` (#345), and `apps/client` (#347) — do NOT reintroduce a per-package copy. The client uses the package's `browser` entry (`@docmost/prosemirror-markdown/browser`): markdown paste (`markdown-clipboard.ts`), copy-as-markdown, and AI-chat rendering now all go through the canonical converter, so the hand-written `marked`/`turndown` markdown layer that used to live in `editor-ext` was deleted (#347). The browser entry runs the HTML→DOM stage on the native `DOMParser`, so jsdom stays out of the client bundle. `editor-ext` is the upstream source of the Tiptap schema; the package's `docmost-schema.ts` mirrors it and a serializer-contract test (`packages/prosemirror-markdown/test/serializer-contract.test.ts`) guards the boundary (every schema node must have a converter case), so a drift surfaces as a failing test rather than silent divergence. For the converter's property-testing and counterexample→fixture process (P1–P4 invariants, the `PROPERTY_SEED`/`PROPERTY_NUM_RUNS` knobs, and the nightly fuzz workflow), see `packages/prosemirror-markdown/README.md`.
 - API access goes through `apps/client/src/lib/api-client.ts` (axios). The `@` alias maps to `apps/client/src`.
 - Runtime config is injected at build time by `vite.config.ts` via `define` (`APP_URL`, `COLLAB_URL`, `APP_VERSION`, …) — these come from the root `.env`, not from `import.meta.env`.
 

CHANGELOG.md

@@ -251,9 +251,37 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
   by physical key position and matched against the commands; genuine Cyrillic
   search terms keep priority over remapped candidates, and short wrong-layout
   prefixes match by command title. (#283, #285, #287)
+- **Opt-in substring "lookup" search mode for agents.** `/api/search` gains an
+  additive, opt-in mode (guarded by a new `substring` flag) that matches literal
+  substrings of page titles and body text — so technical tokens the full-text
+  tokenizer mangles (`backup-srv.local`, `10.0.12.5`, `WB-MGE-30D86B`) are found
+  even when the FTS query is empty. It returns a location `path`, a windowed
+  `snippet` and a per-response relevance `score`, supports `titleOnly` and a
+  `parentPageId` subtree scope, and applies the page-level permission filter
+  before the limit. The web UI never sets `substring`, so its full-text search
+  behaviour is byte-for-byte unchanged. The leading-wildcard `LIKE` predicates
+  are backed by GIN trigram indexes on `LOWER(f_unaccent(title))` and
+  `LOWER(f_unaccent(text_content))` so lookups use a bitmap index scan instead of
+  a sequential scan. (#443)
+- **MCP `search` tool returns richer, agent-oriented results.** The external MCP
+  `search` response shape changes for the agent surface: each hit now carries
+  `pageId` (renamed from `id`), plus `path`, `snippet` and `score`; the
+  UI-oriented `spaceId`, `rank` and `highlight` fields are dropped. (#443)
 
 ### Changed
 
+- **Client markdown paste/copy and AI-chat rendering now go through the canonical
+  converter.** Pasting markdown into the editor, "Copy as markdown", the AI title
+  generator, and the AI-chat markdown renderer all now use
+  `@docmost/prosemirror-markdown` (via its new `browser` entry — native
+  `DOMParser`, no jsdom in the client bundle) instead of the hand-written
+  `marked`/`turndown` markdown layer in `editor-ext`, which was **deleted**. As a
+  result, pasting canonical markdown (`^[…]` footnotes, `<!--img …-->`,
+  `> [!type]` callouts, `$…$` math, `==…==` highlight, standalone `<!--subpages-->`
+  comments) now produces the SAME nodes the server import produces for the same
+  text. Chat/reasoning markdown now renders through the editor schema (list items
+  are wrapped in `<p>`; CSS keeps them tight). (#347)
+
 - **Enabling a public share no longer auto-shares the whole sub-tree.** Turning
   a page "Shared to web" now defaults to the page alone; descendant pages become
   public only when you explicitly turn on the dedicated "Include sub-pages"

Dockerfile

@@ -45,6 +45,11 @@ COPY --from=builder /app/packages/editor-ext/dist /app/packages/editor-ext/dist
 COPY --from=builder /app/packages/editor-ext/package.json /app/packages/editor-ext/package.json
 COPY --from=builder /app/packages/mcp/build /app/packages/mcp/build
 COPY --from=builder /app/packages/mcp/package.json /app/packages/mcp/package.json
+# The mcp package reads its data files (drawio-presets.json, drawio-shape-index.json.gz)
+# at runtime via `new URL("../../data/…", import.meta.url)` relative to build/lib/*.js,
+# i.e. from packages/mcp/data/. tsc emits only build/, so ship data/ explicitly or
+# drawioFromGraph and the shape catalog die with ENOENT on packages/mcp/data/*.
+COPY --from=builder /app/packages/mcp/data /app/packages/mcp/data
 # mcp now depends on @docmost/prosemirror-markdown (workspace:*) and eager-imports
 # it at runtime (the in-app ai-chat DocmostClient loads build/index.js -> lib/
 # markdown-converter.js). Ship the built package + its manifest, or the prod
@@ -81,4 +86,14 @@ VOLUME ["/app/data/storage"]
 
 EXPOSE 3000
 
+# DEPLOY REQUIREMENT — SINGLE INSTANCE or STICKY SESSIONS (#449).
+# MCP content writes are serialized per page by an IN-PROCESS mutex, and the
+# stash_page blob store + cached collab sessions are RAM-only and process-local.
+# Running MULTIPLE replicas of this image behind a load balancer WITHOUT sticky
+# sessions silently breaks per-page write serialization (two replicas can lock
+# the same page at once) and makes stash_page blobs unreachable across replicas.
+# Run a SINGLE instance, or pin each page's traffic to one replica (sticky
+# sessions / consistent hashing on page id). There is deliberately no
+# cross-process lock yet — a conscious constraint. See .env.example (the "MCP
+# collaboration write path" block) and packages/mcp/README.md for details.
 CMD ["pnpm", "start"]

apps/client/package.json

@@ -21,6 +21,7 @@
     "@atlaskit/pragmatic-drag-and-drop-live-region": "1.3.4",
     "@casl/react": "5.0.1",
     "@docmost/editor-ext": "workspace:*",
+    "@docmost/prosemirror-markdown": "workspace:*",
     "@excalidraw/excalidraw": "0.18.0-3a5ef40",
     "@mantine/core": "8.3.18",
     "@mantine/dates": "8.3.18",

apps/client/src/features/ai-chat/components/ai-chat.module.css

@@ -55,6 +55,15 @@
     padding-inline-start: 1.4em;
 }
 
+/* The canonical converter renders list items through the editor schema, which
+   wraps each item's content in a <p> (listItem content is `paragraph+`). Drop
+   that paragraph's block margin so list items render TIGHT (no extra vertical
+   gap), matching the previous marked output — same rule already applied to
+   table cells above (issue #347). */
+.markdown li p {
+    margin: 0;
+}
+
 /* GFM tables in assistant markdown. The chat lives in a NARROW side panel, so a
    wide LLM table must scroll horizontally instead of collapsing its columns:
    `.markdown` sets `word-break: break-word`, which (with the default table
@@ -172,6 +181,14 @@
     margin: 0 0 4px;
 }
 
+/* Same as `.markdown li p` above: the canonical converter wraps every list
+   item's content in a <p>, so without this each reasoning-panel list item would
+   pick up `.reasoningText p`'s 4px bottom margin and render too loose. Drop it
+   so Reasoning-panel lists stay tight, mirroring the pre-#347 marked output. */
+.reasoningText li p {
+    margin: 0;
+}
+
 .inputWrapper {
     flex: 0 0 auto;
     padding-top: var(--mantine-spacing-xs);

apps/client/src/features/ai-chat/utils/collapse-blank-lines.test.ts

@@ -33,29 +33,44 @@ describe("collapseBlankLines", () => {
   });
 });
 
-describe("collapseBlankLines + renderChatMarkdown (tight reasoning rendering)", () => {
+describe("collapseBlankLines + renderChatMarkdown (canonical converter)", () => {
-  it("renders a blank-line-separated list as a TIGHT list (no <li><p>)", () => {
+  // Chat markdown now renders through @docmost/prosemirror-markdown (issue #347):
+  // the SAME converter the editor/import use. Its list items are schema-shaped —
+  // each <li>'s content is wrapped in a <p> (listItem content is `paragraph+`) —
+  // so the HTML always carries `<li><p>…</p></li>` regardless of blank-line
+  // looseness in the source (the converter has no tight/loose distinction). The
+  // visual tightness that `collapseBlankLines` used to buy is now provided by
+  // CSS (`.markdown li p { margin: 0 }`), not the HTML shape.
+  it("renders a blank-line-separated bullet list as a real <ul> list", () => {
     const loose =
       "Intro paragraph.\n\n- item one\n\n- item two\n\n- item three";
     const html = renderChatMarkdown(collapseBlankLines(loose), {});
-    // Tight list: each <li> holds the text directly, not wrapped in a <p>.
+    // Clean, un-namespaced HTML (DOMSerializer, not XMLSerializer) — no xmlns.
-    expect(html).toContain("<li>item one</li>");
-    expect(html).not.toContain("<li><p>");
-    // The list still parses as a list after the paragraph (not a paragraph+<br>).
     expect(html).toContain("<ul>");
+    expect(html).not.toMatch(/<ul[^>]*xmlns/);
+    // The item text is present (inside the schema's <li><p> wrapper).
+    expect(html).toContain("item one");
+    // The intro paragraph renders as its own paragraph before the list.
     expect(html).toContain("<p>Intro paragraph.</p>");
   });
 
-  it("renders an ordered list (1. 2.) as tight after collapsing", () => {
+  it("renders an ordered list (1. 2.) as a real <ol> list", () => {
     const loose = "Intro.\n\n1. first\n\n2. second";
     const html = renderChatMarkdown(collapseBlankLines(loose), {});
     expect(html).toContain("<ol>");
-    expect(html).toContain("<li>first</li>");
+    expect(html).not.toMatch(/<ol[^>]*xmlns/);
-    expect(html).not.toContain("<li><p>");
+    expect(html).toContain("first");
+    expect(html).toContain("second");
   });
 
-  it("the loose source WOULD render <li><p> without collapsing (control)", () => {
+  it("wraps list-item content in <p> (schema shape; tightness is CSS)", () => {
+    // The canonical converter always wraps a list item's content in a paragraph,
+    // whether or not the source had blank lines between items.
     const loose = "- a\n\n- b";
     expect(renderChatMarkdown(loose, {})).toContain("<li><p>");
+    // And a "tight" source produces the identical wrapping (no distinction).
+    expect(renderChatMarkdown(collapseBlankLines(loose), {})).toContain(
+      "<li><p>",
+    );
   });
 });

apps/client/src/features/ai-chat/utils/markdown.ts

@@ -1,6 +1,37 @@
-import { markdownToHtml } from "@docmost/editor-ext";
+import {
+  markdownToProseMirrorSync,
+  docmostExtensions,
+} from "@docmost/prosemirror-markdown/browser";
+import { getSchema } from "@tiptap/core";
+import { Node as PMNode, DOMSerializer } from "@tiptap/pm/model";
 import DOMPurify from "dompurify";
 
+// The Docmost editor schema, built once. Chat markdown is rendered through the
+// SAME schema the editor/import use (issue #347), so chat output matches how the
+// page would render the same markdown.
+const chatSchema = getSchema(docmostExtensions);
+
+/**
+ * Markdown -> HTML for chat display, via the canonical converter. We serialize
+ * the ProseMirror doc with `DOMSerializer` into a real element and read its
+ * `innerHTML` (rather than `@tiptap/html`'s `generateHTML`, whose browser path
+ * uses `XMLSerializer` and stamps a `xmlns` on every block) so the markup is
+ * clean HTML. `li > p` wrapping is inherent to the schema (listItem content is
+ * `paragraph+`); the chat CSS zeroes those paragraph margins so lists still
+ * render tight.
+ */
+function markdownToChatHtml(markdown: string): string {
+  const doc = markdownToProseMirrorSync(markdown);
+  const node = PMNode.fromJSON(chatSchema, doc);
+  const div = document.createElement("div");
+  DOMSerializer.fromSchema(chatSchema).serializeFragment(
+    node.content,
+    { document },
+    div,
+  );
+  return div.innerHTML;
+}
+
 export interface RenderChatMarkdownOptions {
   /**
    * Neutralize INTERNAL links so they render as inert text (no `href`/`target`).
@@ -63,22 +94,32 @@ function neutralizeInternalLinksHook(node: Element): void {
 
 /**
  * Render AI markdown to sanitized HTML for read-only display. We reuse the
- * app's `markdownToHtml` (the same `marked` pipeline used for paste/import) so
+ * canonical converter (issue #347): markdown -> ProseMirror JSON (the SAME
- * chat output matches the editor's markdown flavor, then sanitize with
+ * `markdownToProseMirrorSync` the editor paste/import path uses, so chat output
- * DOMPurify — LLM output is untrusted, so it must never reach the DOM unsanitized.
+ * matches the editor's markdown flavor) -> HTML via `markdownToChatHtml`
+ * (DOMSerializer), then sanitize with DOMPurify — LLM output is untrusted, so it
+ * must never reach the DOM unsanitized.
  *
- * `markdownToHtml` can return `string | Promise<string>` (it has async marked
+ * Stays SYNCHRONOUS: both callers render inside React (a memo and a useMemo),
- * extensions registered). In practice plain chat markdown resolves
+ * so the whole pipeline must resolve without awaiting. The converter's sync
- * synchronously, but we guard the Promise case by returning a safe empty string
+ * entry makes that possible; on any conversion error we return "" so the caller
- * for that branch (the caller renders the raw text fallback instead).
+ * falls back to raw text (the same fallback the old Promise-guard produced).
  */
 export function renderChatMarkdown(
   markdown: string,
   options: RenderChatMarkdownOptions = {},
 ): string {
   if (!markdown) return "";
-  const html = markdownToHtml(markdown);
+  let html: string;
-  if (typeof html !== "string") return "";
+  try {
+    // markdown -> canonical PM JSON -> HTML (native DOMParser in the browser;
+    // jsdom is never bundled — see @docmost/prosemirror-markdown/browser).
+    html = markdownToChatHtml(markdown);
+  } catch {
+    // Malformed/unsupported markdown must not crash the chat render; fall back
+    // to raw text (empty return -> caller shows the plain-text branch).
+    return "";
+  }
 
   if (!options.neutralizeInternalLinks) {
     // Internal chat: unchanged behavior, no hook registered.

apps/client/src/features/editor/extensions/markdown-clipboard.paste.test.ts

@@ -0,0 +1,206 @@
+import { describe, it, expect } from "vitest";
+import { Editor } from "@tiptap/core";
+import { Document } from "@tiptap/extension-document";
+import { Paragraph } from "@tiptap/extension-paragraph";
+import { Text } from "@tiptap/extension-text";
+import { Bold } from "@tiptap/extension-bold";
+import { Italic } from "@tiptap/extension-italic";
+import { MarkdownClipboard } from "./markdown-clipboard";
+
+/**
+ * Integration coverage for the async `handlePaste` seam (issue #347). The paste
+ * conversion moved to `@docmost/prosemirror-markdown`'s browser entry, whose
+ * `markdownToProseMirror` is async — so `handlePaste` captures the range, claims
+ * the event (returns true), and dispatches the insert on the next microtask.
+ * These tests drive that path end to end on a minimal schema (a plain-markdown
+ * paste whose converted nodes fit paragraph/text/bold/italic), asserting the
+ * text lands with the right marks and that the raw markdown syntax is consumed
+ * (recognized as markdown, not inserted literally).
+ */
+
+function makeEditor() {
+  const element = document.createElement("div");
+  document.body.appendChild(element);
+  return new Editor({
+    element,
+    extensions: [
+      Document,
+      Paragraph,
+      Text,
+      Bold,
+      Italic,
+      MarkdownClipboard.configure({ transformPastedText: true }),
+    ],
+    content: { type: "doc", content: [{ type: "paragraph" }] },
+  });
+}
+
+// Locate the markdownClipboard plugin and invoke its handlePaste directly with a
+// synthetic clipboard event (jsdom has no real paste pipeline). The plugin's
+// handlePaste closes over the extension `this`, so calling it off the plugin
+// props preserves `this.editor`/`this.options`.
+function paste(editor: Editor, text: string): boolean {
+  const view = editor.view;
+  const plugin = view.state.plugins.find(
+    (p: any) => p.props && p.spec?.key,
+  ) as any;
+  const event = {
+    clipboardData: {
+      getData: (type: string) => (type === "text/plain" ? text : ""),
+    },
+  } as unknown as ClipboardEvent;
+  // Find the specific handlePaste that belongs to the markdown clipboard plugin.
+  const md = view.state.plugins.find(
+    (p: any) => typeof p.props?.handlePaste === "function",
+  ) as any;
+  return md.props.handlePaste(view, event, view.state.selection.content());
+}
+
+// Flush the microtask queue so the async .then() dispatch runs.
+const flush = () => new Promise((r) => setTimeout(r, 0));
+
+describe("MarkdownClipboard handlePaste (async md -> PM)", () => {
+  it("converts a plain-markdown paste with bold/italic into marked text", async () => {
+    const editor = makeEditor();
+    const claimed = paste(editor, "hello **bold** and *italic*");
+    // The paste is claimed synchronously (async insert follows).
+    expect(claimed).toBe(true);
+    await flush();
+
+    const json = editor.getJSON();
+    const text = JSON.stringify(json);
+    // The raw markdown asterisks are consumed (recognized), not inserted literally.
+    expect(editor.getText()).not.toContain("**");
+    expect(editor.getText()).toContain("bold");
+    expect(editor.getText()).toContain("italic");
+    // The bold/italic marks materialized.
+    expect(text).toContain('"bold"');
+    expect(text).toContain('"italic"');
+    editor.destroy();
+  });
+
+  it("recognizes a bullet list paste as list structure (not literal '-')", async () => {
+    // A bullet list is not representable in this minimal schema, so the converter
+    // output would fail PMNode.fromJSON and the catch inserts raw text. Use a
+    // paste whose nodes DO fit the schema to assert the happy path instead: two
+    // paragraphs separated by a blank line.
+    const editor = makeEditor();
+    paste(editor, "first para\n\nsecond para");
+    await flush();
+    const json = editor.getJSON() as any;
+    const paras = (json.content || []).filter(
+      (n: any) => n.type === "paragraph",
+    );
+    // Two paragraphs materialized from the blank-line-separated markdown.
+    expect(paras.length).toBeGreaterThanOrEqual(2);
+    expect(editor.getText()).toContain("first para");
+    expect(editor.getText()).toContain("second para");
+    editor.destroy();
+  });
+
+  it("falls back to raw text when conversion yields nodes the schema lacks", async () => {
+    // `# heading` converts to a `heading` node absent from this minimal schema,
+    // so PMNode.fromJSON throws and the catch re-inserts the raw text — the user
+    // never loses their clipboard content.
+    const editor = makeEditor();
+    paste(editor, "# a heading line");
+    await flush();
+    // Content is preserved (either as heading text or literal), never dropped.
+    expect(editor.getText()).toContain("a heading line");
+    editor.destroy();
+  });
+});
+
+// The async seam captures the target range synchronously, then replaces on the
+// next microtask. If the document changed under it between capture and resolve
+// (impossible in prod — same microtask — but pinned here), BOTH the success
+// (replaceRange) and the fail-open (insertText) branches must fall back to the
+// LIVE selection rather than a stale absolute range, so neither clobbers content
+// nor throws a RangeError. We force the mid-flight change by dispatching a
+// doc-mutating transaction AFTER the synchronous claim but BEFORE flushing the
+// microtask that runs the `.then`/`.catch`.
+describe("MarkdownClipboard handlePaste — doc-changed-mid-flight guard", () => {
+  // Replace the whole doc with one paragraph of `text` (synchronous dispatch).
+  // An empty string yields an empty paragraph (a text node may not be empty).
+  function seedContent(editor: Editor, text: string) {
+    editor.commands.setContent({
+      type: "doc",
+      content: [
+        text
+          ? { type: "paragraph", content: [{ type: "text", text }] }
+          : { type: "paragraph" },
+      ],
+    });
+  }
+
+  it("success branch: mid-flight doc change routes the paste to the LIVE selection, never the stale range (clobber-proving)", async () => {
+    // The paste captures a NON-EMPTY range {1,5} (over "AAAA"). Then, before the
+    // async resolve, the doc GROWS ("MARKER" inserted at the start) and the cursor
+    // is parked at the doc END. The captured {1,5} is now stale and points INTO
+    // "MARKER". A WORKING guard replaces at the live (end) selection → MARKER is
+    // untouched. A BROKEN guard replaces the stale {1,5} → it erases the first
+    // characters of MARKER (this is what a zero-width `from==to` range could never
+    // reveal, which is why the earlier version was vacuous).
+    const editor = makeEditor();
+    seedContent(editor, "AAAABBBB");
+    editor.commands.setTextSelection({ from: 1, to: 5 }); // captured range = {1,5}
+    const claimed = paste(editor, "hello **bold**");
+    expect(claimed).toBe(true);
+
+    // Mid-flight: grow the doc and move the cursor to a KNOWN-safe end position.
+    editor.view.dispatch(editor.view.state.tr.insertText("MARKER", 1));
+    const end = editor.state.doc.content.size;
+    editor.commands.setTextSelection({ from: end, to: end });
+    await flush();
+
+    const text = editor.getText();
+    // MARKER intact only if the guard used the live selection, not the stale range.
+    expect(text).toContain("MARKER");
+    expect(text).toContain("bold");
+    expect(text).not.toContain("**");
+    editor.destroy();
+  });
+
+  it("fail-open branch: a mid-flight doc SHRINK makes the stale `to` out of bounds — the guard must avoid a RangeError (throw-proving)", async () => {
+    // The paste captures a range {1,9} over an 8-char paragraph, then the
+    // conversion FAILS (`# heading` -> a heading node the minimal schema lacks,
+    // so PMNode.fromJSON throws -> the fail-open catch runs). Before the reject,
+    // the doc is SHRUNK to an empty paragraph, so the captured `to` (9) is now far
+    // past the doc's end. A WORKING guard inserts the raw text at the live (valid)
+    // selection → "raw heading" lands. A BROKEN guard does insertText(md, 1, 9) on
+    // a size-2 doc → RangeError, so the dispatch never runs and "raw heading" is
+    // absent (the assertion reddens). A zero-width/growing-doc setup could never
+    // push `to` out of bounds, which is why the earlier version was vacuous.
+    const editor = makeEditor();
+    seedContent(editor, "AAAABBBB");
+    editor.commands.setTextSelection({ from: 1, to: 9 }); // captured range = {1,9}
+    paste(editor, "# raw heading");
+
+    // Mid-flight: shrink the doc so the captured `to` = 9 is now out of bounds.
+    seedContent(editor, "");
+    await flush();
+
+    const text = editor.getText();
+    // Raw text lands (via the live selection) only if the guard avoided the
+    // stale, now-out-of-bounds range.
+    expect(text).toContain("raw heading");
+    editor.destroy();
+  });
+
+  it("two pastes in flight: neither payload is lost (no data loss)", async () => {
+    // Prod-unreachable (two paste events are separate macrotasks, and each
+    // conversion resolves on a microtask before the next), but pinned here: when
+    // both resolve back-to-back, the second sees the changed doc and inserts at
+    // the live selection the first left — so the two payloads may INTERLEAVE, but
+    // neither is dropped. We assert no data loss, not contiguity.
+    const editor = makeEditor();
+    paste(editor, "alphaword");
+    paste(editor, "betaword");
+    await flush();
+    const text = editor.getText();
+    // Neither payload fully dropped (interleaving may split one of them).
+    expect(text).toContain("alpha");
+    expect(text).toContain("beta");
+    editor.destroy();
+  });
+});

apps/client/src/features/editor/extensions/markdown-clipboard.test.ts

@@ -1,5 +1,12 @@
 import { describe, it, expect } from "vitest";
-import { htmlToMarkdown } from "@docmost/editor-ext";
+// Markdown conversion now goes through the canonical package's BROWSER entry
+// (issue #347): the same converter the server import/export uses, resolved via
+// the `browser` exports condition so it runs on the native `DOMParser` (the
+// client jsdom vitest env provides one) with jsdom never bundled.
+import {
+  convertProseMirrorToMarkdown,
+  markdownToProseMirrorSync,
+} from "@docmost/prosemirror-markdown/browser";
 import {
   normalizeTableColumnWidths,
   classifyClipboardSelection,
@@ -175,10 +182,13 @@ describe("classifyClipboardSelection", () => {
 
 // Output-level tests for the table clipboard regression: copying a table must
 // yield a real GFM pipe table, NOT one-value-per-line concatenated cells.
-// These exercise the actual markdown produced by htmlToMarkdown (the same
+// These exercise the actual markdown produced by convertProseMirrorToMarkdown —
-// serializer step the clipboardTextSerializer runs), so they pin the OUTPUT
+// the same serializer step the clipboardTextSerializer now runs (issue #347) —
-// shape that the classifier-flag tests above do not cover.
+// so they pin the OUTPUT shape that the classifier-flag tests above do not cover.
-describe("table clipboard markdown output (htmlToMarkdown)", () => {
+// Input is ProseMirror JSON (what the copied slice serializes to), matching the
+// clipboardTextSerializer's new call: it wraps the slice content in a synthetic
+// `doc` (and the bare-rows case in a `table`) and calls the converter.
+describe("table clipboard markdown output (convertProseMirrorToMarkdown)", () => {
   // Trim each line and drop blanks so structural assertions are whitespace-robust.
   function lines(md: string): string[] {
     return md
@@ -188,10 +198,10 @@ describe("table clipboard markdown output (htmlToMarkdown)", () => {
   }
 
   // A GFM separator row like "| --- | --- |" (any number of columns), tolerant
-  // of the padding turndown emits.
+  // of the padding the serializer emits.
   function isSeparatorRow(line: string): boolean {
     const compact = line.replace(/\s+/g, "");
-    return /^\|(?:-{3,}\|)+$/.test(compact);
+    return /^\|(?::?-{2,}:?\|)+$/.test(compact);
   }
 
   // Split a pipe-delimited row into trimmed cell values.
@@ -203,42 +213,33 @@ describe("table clipboard markdown output (htmlToMarkdown)", () => {
       .map((c) => c.trim());
   }
 
-  it("serializes a header-less partial cell selection (bare rows) as a valid GFM pipe table", () => {
+  const cell = (t: string) => ({
-    // Mirror the serializer's `wrapBareRows` branch exactly: bare <tr> nodes are
+    type: "tableCell",
-    // wrapped in <table><tbody> and htmlToMarkdown(div.innerHTML) is called.
+    content: [{ type: "paragraph", content: [{ type: "text", text: t }] }],
-    // See markdown-clipboard.ts clipboardTextSerializer:
+  });
-    //   const table = document.createElement("table");
+  const headerCell = (t: string) => ({
-    //   const tbody = document.createElement("tbody");
+    type: "tableHeader",
-    //   tbody.appendChild(fragment); table.appendChild(tbody);
+    content: [{ type: "paragraph", content: [{ type: "text", text: t }] }],
-    //   div.appendChild(table);
+  });
-    //   return htmlToMarkdown(div.innerHTML);
+  const row = (nodes: any[]) => ({ type: "tableRow", content: nodes });
-    const div = document.createElement("div");
-    const table = document.createElement("table");
-    const tbody = document.createElement("tbody");
-    for (const [c1, c2] of [
-      ["a", "b"],
-      ["c", "d"],
-    ]) {
-      const tr = document.createElement("tr");
-      const td1 = document.createElement("td");
-      td1.textContent = c1;
-      const td2 = document.createElement("td");
-      td2.textContent = c2;
-      tr.appendChild(td1);
-      tr.appendChild(td2);
-      tbody.appendChild(tr);
-    }
-    table.appendChild(tbody);
-    div.appendChild(table);
 
-    const md = htmlToMarkdown(div.innerHTML);
+  it("serializes a header-less partial cell selection (bare rows) as a valid GFM pipe table", () => {
+    // Mirror the serializer's `wrapBareRows` branch: bare tableRow nodes are
+    // wrapped in a synthetic `table` and convertProseMirrorToMarkdown is called
+    // (see markdown-clipboard.ts clipboardTextSerializer).
+    const rows = [
+      row([cell("a"), cell("b")]),
+      row([cell("c"), cell("d")]),
+    ];
+    const md = convertProseMirrorToMarkdown({
+      type: "doc",
+      content: [{ type: "table", content: rows }],
+    });
     const ls = lines(md);
 
-    // Valid GFM: a header/data separator row is present (an empty header is
+    // Valid GFM: a header/data separator row is present.
-    // synthesized by the GFM turndown plugin for a header-less table — fine).
     expect(ls.some(isSeparatorRow)).toBe(true);
-    // NOT the old broken "one value per line" shape: every line is pipe-delimited
+    // NOT the old broken "one value per line" shape: every line is pipe-delimited.
-    // and no line is a bare cell value on its own.
     expect(ls.every((l) => l.includes("|"))).toBe(true);
     expect(md).not.toMatch(/^\s*(a|b|c|d)\s*$/m);
     // The cell values land in real pipe-delimited data rows.
@@ -248,39 +249,21 @@ describe("table clipboard markdown output (htmlToMarkdown)", () => {
   });
 
   it("serializes a whole table with a header row as a proper GFM table (headline regression)", () => {
-    // Mirror the serializer's non-wrap branch: the full <table> node is appended
+    // Mirror the serializer's non-wrap branch: the full `table` node is the
-    // directly (div.appendChild(fragment)) and htmlToMarkdown(div.innerHTML) runs.
+    // slice content and convertProseMirrorToMarkdown runs on it.
-    const div = document.createElement("div");
+    const md = convertProseMirrorToMarkdown({
-    const table = document.createElement("table");
+      type: "doc",
-
+      content: [
-    const thead = document.createElement("thead");
+        {
-    const headerRow = document.createElement("tr");
+          type: "table",
-    for (const h of ["Name", "Age"]) {
+          content: [
-      const th = document.createElement("th");
+            row([headerCell("Name"), headerCell("Age")]),
-      th.textContent = h;
+            row([cell("Alice"), cell("30")]),
-      headerRow.appendChild(th);
+            row([cell("Bob"), cell("25")]),
-    }
+          ],
-    thead.appendChild(headerRow);
+        },
-    table.appendChild(thead);
+      ],
-
+    });
-    const tbody = document.createElement("tbody");
-    for (const [name, age] of [
-      ["Alice", "30"],
-      ["Bob", "25"],
-    ]) {
-      const tr = document.createElement("tr");
-      const td1 = document.createElement("td");
-      td1.textContent = name;
-      const td2 = document.createElement("td");
-      td2.textContent = age;
-      tr.appendChild(td1);
-      tr.appendChild(td2);
-      tbody.appendChild(tr);
-    }
-    table.appendChild(tbody);
-    div.appendChild(table);
-
-    const md = htmlToMarkdown(div.innerHTML);
     const ls = lines(md);
 
     // Proper GFM structure: separator row + all rows pipe-delimited.
@@ -296,3 +279,146 @@ describe("table clipboard markdown output (htmlToMarkdown)", () => {
     expect(md).not.toMatch(/^\s*(Name|Age|Alice|Bob|30|25)\s*$/m);
   });
 });
+
+// #347 acceptance: pasting CANONICAL markdown yields the SAME nodes the server
+// import produces for the same text. The paste path calls markdownToProseMirror
+// (the package browser entry) — the identical converter the server import uses —
+// so asserting the converter (via the browser entry, on the native DOMParser)
+// recognizes each canon form pins the paste-parity guarantee. These forms were
+// NOT recognized by the old editor-ext marked layer the paste used before.
+describe("canonical markdown paste recognition (browser entry parity)", () => {
+  // Collect every node type present in a doc (recursively).
+  const collectTypes = (n: any, set = new Set<string>()): Set<string> => {
+    if (!n || typeof n !== "object") return set;
+    if (n.type) set.add(n.type);
+    if (Array.isArray(n.content)) n.content.forEach((c) => collectTypes(c, set));
+    return set;
+  };
+  const findNode = (n: any, type: string): any => {
+    if (!n || typeof n !== "object") return undefined;
+    if (n.type === type) return n;
+    if (Array.isArray(n.content)) {
+      for (const c of n.content) {
+        const hit = findNode(c, type);
+        if (hit) return hit;
+      }
+    }
+    return undefined;
+  };
+  const allText = (n: any): string => {
+    if (!n || typeof n !== "object") return "";
+    if (typeof n.text === "string") return n.text;
+    if (Array.isArray(n.content)) return n.content.map(allText).join("");
+    return "";
+  };
+
+  it("^[…] inline footnote -> footnoteReference + footnotesList", () => {
+    const doc = markdownToProseMirrorSync("Body^[a note here].");
+    const types = collectTypes(doc);
+    expect(types.has("footnoteReference")).toBe(true);
+    expect(types.has("footnotesList")).toBe(true);
+    expect(types.has("footnoteDefinition")).toBe(true);
+  });
+
+  it('<!--img {…}--> attached image comment -> image with align', () => {
+    const doc = markdownToProseMirrorSync(
+      '![alt](/files/x.png) <!--img {"align":"left"}-->',
+    );
+    const img = findNode(doc, "image");
+    expect(img).toBeTruthy();
+    expect(img.attrs?.align).toBe("left");
+    expect(img.attrs?.src).toBe("/files/x.png");
+  });
+
+  it("> [!type] Obsidian callout -> callout node with type", () => {
+    const doc = markdownToProseMirrorSync("> [!warning]\n> be careful");
+    const callout = findNode(doc, "callout");
+    expect(callout).toBeTruthy();
+    expect(callout.attrs?.type).toBe("warning");
+    expect(allText(callout)).toContain("be careful");
+  });
+
+  it("$…$ inline math -> mathInline node", () => {
+    const doc = markdownToProseMirrorSync("Euler: $e^{i\\pi}+1=0$ done");
+    const math = findNode(doc, "mathInline");
+    expect(math).toBeTruthy();
+    expect(math.attrs?.text).toContain("e^{i\\pi}");
+  });
+
+  it("==…== highlight -> highlight mark", () => {
+    const doc = markdownToProseMirrorSync("A ==marked== word");
+    const marked = findNode(doc, "text");
+    // The highlighted run carries a `highlight` mark somewhere in the doc.
+    const hasHighlight = (n: any): boolean => {
+      if (!n || typeof n !== "object") return false;
+      if (
+        n.type === "text" &&
+        (n.marks || []).some((m: any) => m.type === "highlight")
+      )
+        return true;
+      return Array.isArray(n.content) ? n.content.some(hasHighlight) : false;
+    };
+    expect(marked).toBeTruthy();
+    expect(hasHighlight(doc)).toBe(true);
+  });
+
+  it("<!--subpages--> standalone comment -> subpages node", () => {
+    const doc = markdownToProseMirrorSync("intro\n\n<!--subpages-->\n\nafter");
+    expect(collectTypes(doc).has("subpages")).toBe(true);
+  });
+});
+
+// #347 negatives: plain text carrying markdown-LIKE punctuation must NOT be
+// silently converted/mangled (currency, bare `==`, a `[^1]` reference form).
+describe("plain-text paste negatives (no phantom conversion)", () => {
+  const findNode = (n: any, type: string): any => {
+    if (!n || typeof n !== "object") return undefined;
+    if (n.type === type) return n;
+    if (Array.isArray(n.content)) {
+      for (const c of n.content) {
+        const hit = findNode(c, type);
+        if (hit) return hit;
+      }
+    }
+    return undefined;
+  };
+  const collectTypes = (n: any, set = new Set<string>()): Set<string> => {
+    if (!n || typeof n !== "object") return set;
+    if (n.type) set.add(n.type);
+    if (Array.isArray(n.content)) n.content.forEach((c) => collectTypes(c, set));
+    return set;
+  };
+  const allText = (n: any): string => {
+    if (!n || typeof n !== "object") return "";
+    if (typeof n.text === "string") return n.text;
+    if (Array.isArray(n.content)) return n.content.map(allText).join("");
+    return "";
+  };
+
+  it("currency `$5 and $10` is NOT turned into math", () => {
+    const doc = markdownToProseMirrorSync("It costs $5 and $10 total");
+    expect(findNode(doc, "mathInline")).toBeFalsy();
+    expect(allText(doc)).toContain("$5 and $10");
+  });
+
+  it("a lone `==` is NOT turned into a highlight", () => {
+    const doc = markdownToProseMirrorSync("compare a == b in code");
+    const hasHighlight = (n: any): boolean => {
+      if (!n || typeof n !== "object") return false;
+      if (
+        n.type === "text" &&
+        (n.marks || []).some((m: any) => m.type === "highlight")
+      )
+        return true;
+      return Array.isArray(n.content) ? n.content.some(hasHighlight) : false;
+    };
+    expect(hasHighlight(doc)).toBe(false);
+    expect(allText(doc)).toContain("== b");
+  });
+
+  it("a `[^1]` reference form (no `^[`) is NOT turned into a footnote", () => {
+    const doc = markdownToProseMirrorSync("see note [^1] for details");
+    expect(collectTypes(doc).has("footnoteReference")).toBe(false);
+    expect(allText(doc)).toContain("[^1]");
+  });
+});

apps/client/src/features/editor/extensions/markdown-clipboard.ts

@@ -1,15 +1,23 @@
 // adapted from: https://github.com/aguingand/tiptap-markdown/blob/main/src/extensions/tiptap/clipboard.js - MIT
 import { Extension } from "@tiptap/core";
 import { Plugin, PluginKey, TextSelection } from "@tiptap/pm/state";
-import { DOMParser, DOMSerializer, Fragment, Slice } from "@tiptap/pm/model";
+import { DOMParser, DOMSerializer, Fragment, Slice, Node as PMNode } from "@tiptap/pm/model";
 import { find } from "linkifyjs";
 import {
-  markdownToHtml,
-  htmlToMarkdown,
   canonicalizeFootnotes,
   FOOTNOTES_LIST_NAME,
   FOOTNOTE_REFERENCE_NAME,
 } from "@docmost/editor-ext";
+// Markdown <-> ProseMirror conversion now lives ONLY in the canonical
+// `@docmost/prosemirror-markdown` package (issue #347). The BROWSER entry uses
+// the native `DOMParser` for its HTML->DOM stage (jsdom stays out of the client
+// bundle) while producing the SAME nodes the server import does — so a paste of
+// canonical markdown (`^[…]`, `<!--img …-->`, `> [!type]`, `$…$`, `==…==`,
+// standalone comments) is recognized identically to import.
+import {
+  markdownToProseMirror,
+  convertProseMirrorToMarkdown,
+} from "@docmost/prosemirror-markdown/browser";
 import type { Schema } from "@tiptap/pm/model";
 
 export const MarkdownClipboard = Extension.create({
@@ -39,25 +47,24 @@ export const MarkdownClipboard = Extension.create({
               classifyClipboardSelection(topLevelNodes);
             if (!asMarkdown) return null;
 
-            const div = document.createElement("div");
+            // Convert the copied selection to Markdown through the canonical
-            const serializer = DOMSerializer.fromSchema(this.editor.schema);
+            // package (issue #347), the SAME serializer the server export uses,
-            const fragment = serializer.serializeFragment(slice.content);
+            // so a copied table/list matches the on-disk markdown form. The
-
+            // converter takes a ProseMirror `doc` JSON, so wrap the slice's
+            // top-level content in a synthetic doc.
+            const content = slice.content.toJSON() as any[];
             if (wrapBareRows) {
-              // A partial table cell-selection serializes to bare <tr> nodes
+              // A partial table cell-selection serializes to bare `tableRow`
-              // (prosemirror-tables returns the whole `table` node only when the
+              // nodes (prosemirror-tables yields the whole `table` node only for
-              // entire table is selected). Bare <tr> would be foster-parented
+              // a full-table selection). The converter's table case expects a
-              // away by the HTML parser inside htmlToMarkdown, so wrap them in
+              // `table` wrapper, so wrap the bare rows in one — mirroring the old
-              // <table><tbody> first for the GFM turndown rule to detect them.
+              // <table><tbody> wrap that the HTML->markdown step needed.
-              const table = document.createElement("table");
+              return convertProseMirrorToMarkdown({
-              const tbody = document.createElement("tbody");
+                type: "doc",
-              tbody.appendChild(fragment);
+                content: [{ type: "table", content }],
-              table.appendChild(tbody);
+              });
-              div.appendChild(table);
-            } else {
-              div.appendChild(fragment);
             }
-            return htmlToMarkdown(div.innerHTML);
+            return convertProseMirrorToMarkdown({ type: "doc", content });
           },
           handlePaste: (view, event, slice) => {
             if (!event.clipboardData) {
@@ -95,37 +102,115 @@ export const MarkdownClipboard = Extension.create({
               }
             }
 
-            const { tr } = view.state;
+            const schema = this.editor.schema;
-            const { from, to } = view.state.selection;
+            // Capture the target range NOW. markdownToProseMirror RETURNS A
+            // PROMISE (kept async only for the Node consumers' contract; the
+            // conversion pipeline itself is synchronous), so the actual replace
+            // happens on the next microtask. No user input can interleave a
+            // microtask, so the state is unchanged when we dispatch — but we
+            // still re-read the live state before replacing and, if the doc did
+            // change under us, fall back to the live selection rather than the
+            // captured (now-stale) range.
+            const from = view.state.selection.from;
+            const to = view.state.selection.to;
+            const startDoc = view.state.doc;
+            const md = text.replace(/\n+$/, "");
 
-            const parsed = markdownToHtml(text.replace(/\n+$/, ""));
+            void markdownToProseMirror(md)
-            const body = elementFromString(parsed);
+              .then((doc) => {
-            normalizeTableColumnWidths(body);
+                if (view.isDestroyed) return;
+                // Canonical PM-JSON -> HTML via the LIVE editor schema, then
+                // reuse the UNCHANGED downstream seam (normalizeTableColumnWidths
+                // + parseSlice + canonicalizePastedFootnotes). The JSON->HTML->
+                // JSON hop is lossless (same schema both directions); it lets the
+                // existing paste-insertion logic stay byte-identical — only the
+                // SOURCE of the markdown conversion changed (issue #347 guardrail:
+                // no converter logic in the client, only a call into the package).
+                const node = PMNode.fromJSON(schema, doc);
+                const div = document.createElement("div");
+                DOMSerializer.fromSchema(schema).serializeFragment(
+                  node.content,
+                  { document },
+                  div,
+                );
 
-            const parsedSlice = DOMParser.fromSchema(
+                const body = elementFromString(div.innerHTML);
-              this.editor.schema,
+                normalizeTableColumnWidths(body);
-            ).parseSlice(body, {
-              preserveWhitespace: true,
-            });
 
-            // A markdown paste builds its ProseMirror fragment directly (DOM ->
+                const parsedSlice = DOMParser.fromSchema(schema).parseSlice(
-            // parseSlice), bypassing the editor's footnoteSyncPlugin, which never
+                  body,
-            // reorders an existing list. So a pasted markdown block whose footnote
+                  { preserveWhitespace: true },
-            // definitions are out of order (or contains orphan defs) would be
+                );
-            // stored out of order. Canonicalize the self-contained pasted block so
-            // its footnotes come out reference-ordered, deduped and orphan-free
-            // (issue #228). See canonicalizePastedFootnotes for why this is scoped
-            // to whole-block pastes that carry their own footnotesList.
-            const contentNodes = canonicalizePastedFootnotes(
-              parsedSlice,
-              this.editor.schema,
-            );
 
-            tr.replaceRange(from, to, contentNodes);
+                // A markdown paste builds its ProseMirror fragment directly (DOM
-            const insertEnd = tr.mapping.map(from, 1);
+                // -> parseSlice), bypassing the editor's footnoteSyncPlugin, which
-            tr.setSelection(TextSelection.near(tr.doc.resolve(Math.max(from, insertEnd - 2)), -1));
+                // never reorders an existing list. So a pasted markdown block whose
-            tr.setMeta('paste', true)
+                // footnote definitions are out of order (or contains orphan defs)
-            view.dispatch(tr);
+                // would be stored out of order. Canonicalize the self-contained
+                // pasted block so its footnotes come out reference-ordered, deduped
+                // and orphan-free (issue #228). See canonicalizePastedFootnotes for
+                // why this is scoped to whole-block pastes that carry their own
+                // footnotesList.
+                const contentNodes = canonicalizePastedFootnotes(
+                  parsedSlice,
+                  schema,
+                );
+
+                // Target the captured range (normally still valid — same
+                // microtask). If the doc changed under us since capture, the
+                // captured absolute from/to are stale, so fall back to the live
+                // selection rather than StepMap-mapping the old range.
+                const tr = view.state.tr;
+                let mappedFrom = from;
+                let mappedTo = to;
+                if (view.state.doc !== startDoc) {
+                  // Defensive: if the doc changed under us, fall back to the
+                  // current selection rather than a stale absolute range.
+                  mappedFrom = view.state.selection.from;
+                  mappedTo = view.state.selection.to;
+                }
+                tr.replaceRange(mappedFrom, mappedTo, contentNodes);
+                const insertEnd = tr.mapping.map(mappedFrom, 1);
+                tr.setSelection(
+                  TextSelection.near(
+                    tr.doc.resolve(Math.max(mappedFrom, insertEnd - 2)),
+                    -1,
+                  ),
+                );
+                tr.setMeta("paste", true);
+                view.dispatch(tr);
+              })
+              .catch((err) => {
+                // Fail-open: a conversion error must not swallow the paste
+                // silently in a way that loses the text. We already claimed the
+                // event (returned true), so re-insert the raw text as a plain
+                // paragraph so the user never loses their clipboard content.
+                // Log it: this catch covers BOTH the converter and the success
+                // `.then` body (e.g. PMNode.fromJSON throwing on a schema drift
+                // between the canonical package and the live editor schema), so a
+                // silent degrade to raw text would otherwise be an invisible,
+                // non-reproducible regression ("my table pasted as text").
+                console.error(
+                  "markdown paste conversion failed, inserting raw text",
+                  err,
+                );
+                if (view.isDestroyed) return;
+                const tr = view.state.tr;
+                // Same guard the success path uses: if the doc changed under us
+                // since the range was captured (normally never — same microtask),
+                // the captured absolute from/to are stale and would throw a
+                // RangeError here (an unhandled rejection on a hot paste path).
+                // Fall back to the live selection instead of a stale range.
+                if (view.state.doc !== startDoc) {
+                  const sel = view.state.selection;
+                  tr.insertText(md, sel.from, sel.to);
+                } else {
+                  tr.insertText(md, from, to);
+                }
+                tr.setMeta("paste", true);
+                view.dispatch(tr);
+              });
+            // Claim the paste: we insert asynchronously above.
             return true;
           },
           // Strip trailing whitespace-only paragraphs from pasted content.

apps/client/src/features/editor/hooks/use-generate-page-title.test.tsx

@@ -33,10 +33,11 @@ vi.mock("@/lib/local-emitter.ts", () => ({
   default: { emit: (...args: unknown[]) => localEmitMock(...args) },
 }));
 
-// htmlToMarkdown just echoes the editor HTML so each test controls the markdown
+// convertProseMirrorToMarkdown echoes a marker carried on the fake editor's
-// purely via the fake page editor's getHTML().
+// getJSON() doc, so each test controls the markdown purely via the fake page
-vi.mock("@docmost/editor-ext", () => ({
+// editor (issue #347: the hook now serializes editor JSON through the package).
-  htmlToMarkdown: (html: string) => html,
+vi.mock("@docmost/prosemirror-markdown/browser", () => ({
+  convertProseMirrorToMarkdown: (doc: { __md?: string }) => doc?.__md ?? "",
 }));
 
 const notificationsShowMock = vi.fn();
@@ -53,10 +54,12 @@ import { useGeneratePageTitle } from "./use-generate-page-title.ts";
 
 // --- Test helpers -------------------------------------------------------------
 
-function makePageEditor(pageId: string, html = "<p>content</p>"): Editor {
+function makePageEditor(pageId: string, md = "content"): Editor {
   return {
     isDestroyed: false,
-    getHTML: () => html,
+    // The mocked convertProseMirrorToMarkdown reads `__md` back off this doc,
+    // so `md` is exactly the markdown the hook will send to the title service.
+    getJSON: () => ({ type: "doc", __md: md }),
     storage: { pageId },
   } as unknown as Editor;
 }

apps/client/src/features/editor/hooks/use-generate-page-title.ts

@@ -3,7 +3,7 @@ import { useMutation } from "@tanstack/react-query";
 import { useAtomValue } from "jotai";
 import { notifications } from "@mantine/notifications";
 import { useTranslation } from "react-i18next";
-import { htmlToMarkdown } from "@docmost/editor-ext";
+import { convertProseMirrorToMarkdown } from "@docmost/prosemirror-markdown/browser";
 import {
   pageEditorAtom,
   titleEditorAtom,
@@ -49,7 +49,9 @@ export function useGeneratePageTitle(pageId: string) {
     mutationFn: async () => {
       if (!pageEditor || pageEditor.isDestroyed) return;
 
-      const markdown = htmlToMarkdown(pageEditor.getHTML()).trim();
+      // Serialize the live editor content to markdown through the canonical
+      // converter (issue #347), matching the on-disk/export markdown form.
+      const markdown = convertProseMirrorToMarkdown(pageEditor.getJSON()).trim();
       if (!markdown) {
         notifications.show({ message: t("The note is empty"), color: "yellow" });
         return;

apps/client/src/features/page/components/header/page-header-menu.tsx

@@ -37,7 +37,7 @@ import { useTreeMutation } from "@/features/page/tree/hooks/use-tree-mutation.ts
 import { PageWidthToggle } from "@/features/user/components/page-width-pref.tsx";
 import { Trans, useTranslation } from "react-i18next";
 import ExportModal from "@/components/common/export-modal";
-import { htmlToMarkdown } from "@docmost/editor-ext";
+import { convertProseMirrorToMarkdown } from "@docmost/prosemirror-markdown/browser";
 import {
   pageEditorAtom,
   yjsConnectionStatusAtom,
@@ -199,8 +199,9 @@ function PageActionMenu({ readOnly }: PageActionMenuProps) {
 
   const handleCopyAsMarkdown = () => {
     if (!pageEditor) return;
-    const html = pageEditor.getHTML();
+    // Copy the page as canonical markdown through the shared converter (issue
-    const markdown = htmlToMarkdown(html);
+    // #347), so "Copy as markdown" matches the server export byte-for-byte.
+    const markdown = convertProseMirrorToMarkdown(pageEditor.getJSON());
     const title = page?.title ? `# ${page.title}\n\n` : "";
     clipboard.copy(`${title}${markdown}`);
     notifications.show({ message: t("Copied") });

apps/server/src/common/helpers/prosemirror/html-embed-import-detect.spec.ts

@@ -1,4 +1,5 @@
-import { markdownToHtml, encodeHtmlEmbedSource } from '@docmost/editor-ext';
+import { markdownToProseMirror } from '@docmost/prosemirror-markdown';
+import { encodeHtmlEmbedSource } from '@docmost/editor-ext';
 import { htmlToJson } from '../../../collaboration/collaboration.util';
 import { hasHtmlEmbedNode, stripHtmlEmbedNodes } from './html-embed.util';
 
@@ -10,13 +11,12 @@ import { hasHtmlEmbedNode, stripHtmlEmbedNodes } from './html-embed.util';
  *
  * The block renders inside a sandboxed iframe, so this is not an XSS surface;
  * this exercises the REAL server import conversion path that ImportService uses
- * (`markdownToHtml` then `htmlToJson`; `processHTML` adds only a cheerio
+ * (`markdownToProseMirror`, the canonical converter — issue #345/#347) and
- * link/iframe normalize pass which does not touch htmlEmbed divs) and asserts
+ * asserts that such a node is DETECTED and STRIPPABLE — so the share read path's
- * that such a node is DETECTED and STRIPPABLE — so the share read path's
  * master-toggle strip can remove it when the workspace toggle is OFF.
  */
 describe('htmlEmbed smuggled via the raw serialized div in imported markdown/HTML', () => {
-  it('round-trips through markdownToHtml -> htmlToJson and is DETECTED (base64 data-source)', async () => {
+  it('round-trips through markdownToProseMirror and is DETECTED (base64 data-source)', async () => {
     const source = '<script>steal()</script>';
     const encoded = encodeHtmlEmbedSource(source);
     const md = [
@@ -27,12 +27,9 @@ describe('htmlEmbed smuggled via the raw serialized div in imported markdown/HTM
       'World',
     ].join('\n');
 
-    const html = await markdownToHtml(md);
+    // The canonical importer parses the raw block-level div into a real
-    // marked preserves the raw block-level div verbatim.
+    // htmlEmbed node carrying the decoded source.
-    expect(html).toContain('data-type="htmlEmbed"');
+    const json = await markdownToProseMirror(md);
-
-    const json = htmlToJson(html);
-    // The div parses into a real htmlEmbed node carrying the decoded source.
     expect(hasHtmlEmbedNode(json)).toBe(true);
 
     // Because it is detected, the share master-toggle strip can remove it.
@@ -59,8 +56,7 @@ describe('htmlEmbed smuggled via the raw serialized div in imported markdown/HTM
     // therefore stripping) does not depend on the source being well-formed, so
     // the bypass cannot be hidden by sending a malformed data-source.
     const md = `<div data-type="htmlEmbed" data-source="&lt;script&gt;x&lt;/script&gt;"></div>`;
-    const html = await markdownToHtml(md);
+    const json = await markdownToProseMirror(md);
-    const json = htmlToJson(html);
     expect(hasHtmlEmbedNode(json)).toBe(true);
     expect(hasHtmlEmbedNode(stripHtmlEmbedNodes(json))).toBe(false);
   });

apps/server/src/core/ai-chat/tools/ai-chat-tools.service.ts

@@ -59,6 +59,8 @@ function __assertClientCallContract(client: DocmostClientLike): void {
   void client.getWorkspace();
   void client.getSpaces();
   void client.listPages(s, n, true);
+  void client.getTree(s, s, n);
+  void client.getPageContext(s);
   void client.listSidebarPages(s, s);
   void client.getOutline(s);
   void client.getPageJson(s);
@@ -121,6 +123,23 @@ function __assertClientCallContract(client: DocmostClientLike): void {
   void client.drawioGet(s, s, 'xml');
   void client.drawioCreate(s, { position: 'append', anchorNodeId: s }, s, s, 'elk');
   void client.drawioUpdate(s, s, s, s, 'elk');
+  // --- draw.io high-level semantic tools (#425 stage 3) ---
+  void client.drawioEditCells(s, s, [{ op: 'delete', cellId: s }], s);
+  void client.drawioFromGraph(
+    s,
+    { position: 'append', anchorNodeId: s },
+    { nodes: [{ id: s, label: s }] },
+    'LR',
+    s,
+    'full',
+    s,
+  );
+  void client.drawioFromMermaid(
+    s,
+    { position: 'append', anchorNodeId: s },
+    s,
+    s,
+  );
   // --- write (comment) ---
   void client.createComment(s, s, 'inline', s, s, s);
   void client.resolveComment(s, true);

apps/server/src/core/ai-chat/tools/docmost-client.loader.ts

@@ -23,6 +23,8 @@ type DocmostClientMethod =
   | 'getWorkspace'
   | 'getSpaces'
   | 'listPages'
+  | 'getTree'
+  | 'getPageContext'
   | 'listSidebarPages'
   | 'getOutline'
   | 'getPageJson'
@@ -69,6 +71,10 @@ type DocmostClientMethod =
   | 'drawioGet'
   | 'drawioCreate'
   | 'drawioUpdate'
+  // --- draw.io high-level semantic tools (#425 stage 3) ---
+  | 'drawioEditCells'
+  | 'drawioFromGraph'
+  | 'drawioFromMermaid'
   // --- write (comment) ---
   | 'createComment'
   | 'resolveComment';

apps/server/src/core/ai-chat/tools/tool-tiers.spec.ts

@@ -27,10 +27,12 @@ import type { DocmostClientLike } from './docmost-client.loader';
  */
 
 describe('tool tier metadata (#332)', () => {
-  it('core set is the documented 13 + searchInPage + insertFootnote (15)', () => {
+  it('core set is the documented 13 + searchInPage + insertFootnote + getTree + getPageContext (17, #443)', () => {
-    expect(CORE_TOOL_KEYS).toHaveLength(15);
+    expect(CORE_TOOL_KEYS).toHaveLength(17);
     expect(CORE_TOOL_SET.has('searchInPage')).toBe(true); // #330, promoted to core
     expect(CORE_TOOL_SET.has('insertFootnote')).toBe(true); // #410, promoted to core
+    expect(CORE_TOOL_SET.has('getTree')).toBe(true); // #443, promoted to core
+    expect(CORE_TOOL_SET.has('getPageContext')).toBe(true); // #443, promoted to core
     // loadTools is a meta-tool, not a normal core key.
     expect(CORE_TOOL_SET.has(LOAD_TOOLS_NAME)).toBe(false);
   });

apps/server/src/core/ai-chat/tools/tool-tiers.ts

@@ -39,12 +39,14 @@ export interface ToolCatalogEntry {
 
 /**
  * CORE (always-active) in-app tool keys — 13 frequent/tiny tools + `searchInPage`
- * (#330) + `insertFootnote` (#410). `searchInPage` is core because it is frequent
+ * (#330) + `insertFootnote` (#410) + `getTree`/`getPageContext` (#443).
- * for the editorial roles this feature targets; `insertFootnote` is core so the
+ * `searchInPage` is core because it is frequent for the editorial roles this
- * footnote tool is NOT hidden while its natural sibling `editPageText` is always
+ * feature targets; `insertFootnote` is core so the footnote tool is NOT hidden
- * active (that asymmetry is exactly what pushed the agent to write literal
+ * while its natural sibling `editPageText` is always active (that asymmetry is
- * `^[...]`). `loadTools` is active too but is not a normal tool key (it is added
+ * exactly what pushed the agent to write literal `^[...]`). `getTree` and
- * to activeTools separately).
+ * `getPageContext` are the single-call navigation/lookup tools — core so the
+ * agent never has to loadTools just to orient itself. `loadTools` is active too
+ * but is not a normal tool key (it is added to activeTools separately).
  */
 export const CORE_TOOL_KEYS = [
   'searchPages',
@@ -66,6 +68,11 @@ export const CORE_TOOL_KEYS = [
   // #410 insertFootnote — core so pinpoint citations to already-written text
   // don't degrade into literal `^[...]`; kept symmetric with editPageText.
   'insertFootnote',
+  // #443 getTree + getPageContext — cheap single-call navigation/lookup tools
+  // (the core listPages even points to getTree); core so the agent never has
+  // to loadTools just to orient itself.
+  'getTree',
+  'getPageContext',
 ] as const;
 
 /** O(1) membership test for the core tier. */

apps/server/src/core/search/dto/search-response.dto.ts

@@ -12,3 +12,22 @@ export class SearchResponseDto {
   updatedAt: Date;
   space: Partial<Space>;
 }
+
+// Response shape for the opt-in agent-lookup mode (#443, `substring: true`).
+// Additive to the FTS response: carries the location (`path`), a windowed
+// `snippet` around the first match and a per-response sort `score`. The MCP
+// layer maps `id → pageId`; `slugId` is never exposed.
+export class SearchLookupResponseDto {
+  id: string;
+  slugId: string;
+  title: string;
+  parentPageId: string | null;
+  // Ancestor titles from the space root down to the direct parent; [] for a
+  // root page.
+  path: string[];
+  // ~300–500 chars around the first match (or a leading text window / extended
+  // ts_headline fallback).
+  snippet: string;
+  // 0..1 float, meaningful ONLY for sorting within one response.
+  score: number;
+}

apps/server/src/core/search/dto/search.dto.ts

@@ -30,6 +30,31 @@ export class SearchDTO {
   @IsOptional()
   @IsNumber()
   offset?: number;
+
+  // --- Opt-in agent-lookup mode (#443). ------------------------------------
+  // These fields are ADDITIVE and default-off: a web client that sends none of
+  // them gets byte-identical FTS behaviour and result shape. They are only read
+  // by the substring/path/snippet code path in SearchService.searchPage.
+  //
+  // NOTE (standalone stdio vs stock upstream): stock upstream validates this DTO
+  // with `whitelist: true`, so an older server silently strips these unknown
+  // fields and the request degrades gracefully to the plain FTS behaviour.
+
+  // Enables the hybrid substring branch (title + text_content LIKE) merged with
+  // the existing FTS branch, plus tiered ranking, path and windowed snippet.
+  @IsOptional()
+  @IsBoolean()
+  substring?: boolean;
+
+  // Restrict the search to a page and all of its descendants (inclusive).
+  @IsOptional()
+  @IsString()
+  parentPageId?: string;
+
+  // Match titles only; do not scan text_content.
+  @IsOptional()
+  @IsBoolean()
+  titleOnly?: boolean;
 }
 
 export class SearchShareDTO extends SearchDTO {

apps/server/src/core/search/search.controller.ts

@@ -60,6 +60,12 @@ export class SearchController {
       }
     }
 
+    // #443 graceful degradation: on EE/Typesense instances the request routes to
+    // the Typesense backend, which does NOT implement the opt-in agent-lookup
+    // mode. The `substring`/`parentPageId`/`titleOnly` fields are silently ignored
+    // and the response carries no `path`/`snippet`/`score` and no substring/tier
+    // ranking — it degrades to plain Typesense FTS. The native lookup mode below
+    // is Postgres-search-driver only.
     if (this.environmentService.getSearchDriver() === 'typesense') {
       return this.searchTypesense(searchDto, {
         userId: user.id,

apps/server/src/core/search/search.service.lookup.spec.ts

@@ -0,0 +1,95 @@
+import {
+  computeLookupScore,
+  escapeLikePattern,
+  SearchLookupTier,
+} from './search.service';
+
+/**
+ * Pure-function coverage for the #443 agent-lookup helpers:
+ *  - escapeLikePattern: LIKE-metacharacter escaping so `%`/`_`/`\` are literals
+ *    (the acceptance-table requirement that a query of `%` or `_` does NOT match
+ *    everything);
+ *  - computeLookupScore: the tiered 0..1 ranking score, where a stronger tier
+ *    always outranks a weaker one regardless of the in-tier secondary signal.
+ *
+ * The DB-touching branch (substring UNION FTS, path CTE, snippet window) is
+ * covered by the integration spec against the real schema.
+ */
+describe('escapeLikePattern', () => {
+  it('escapes the LIKE metacharacters % _ and \\', () => {
+    expect(escapeLikePattern('%')).toBe('\\%');
+    expect(escapeLikePattern('_')).toBe('\\_');
+    expect(escapeLikePattern('\\')).toBe('\\\\');
+  });
+
+  it('escapes the backslash FIRST so it does not double-escape %/_', () => {
+    // Input `\%` must become `\\` + `\%` = `\\\%`, not `\\%`.
+    expect(escapeLikePattern('\\%')).toBe('\\\\\\%');
+  });
+
+  it('leaves ordinary technical chars (. - / digits) untouched', () => {
+    expect(escapeLikePattern('backup-srv.local')).toBe('backup-srv.local');
+    expect(escapeLikePattern('10.0.12')).toBe('10.0.12');
+    expect(escapeLikePattern('WB-MGE-30D86B')).toBe('WB-MGE-30D86B');
+    expect(escapeLikePattern('a/b')).toBe('a/b');
+  });
+
+  it('escapes only the metacharacters in a mixed string', () => {
+    expect(escapeLikePattern('50%_off.zip')).toBe('50\\%\\_off.zip');
+  });
+
+  it('is null/undefined-safe', () => {
+    expect(escapeLikePattern(undefined as any)).toBe('');
+    expect(escapeLikePattern(null as any)).toBe('');
+  });
+});
+
+describe('computeLookupScore', () => {
+  it('keeps every score within (0, 1]', () => {
+    for (const tier of [
+      SearchLookupTier.TITLE_EXACT,
+      SearchLookupTier.TITLE_SUBSTRING,
+      SearchLookupTier.TEXT,
+    ]) {
+      for (const secondary of [0, 0.001, 1, 100, 1e6]) {
+        const s = computeLookupScore({ tier, secondary });
+        expect(s).toBeGreaterThan(0);
+        expect(s).toBeLessThanOrEqual(1);
+      }
+    }
+  });
+
+  it('a stronger tier ALWAYS outranks a weaker tier, whatever the secondary', () => {
+    // Weak tier with a huge secondary must still lose to a strong tier with a
+    // tiny secondary — tiers dominate.
+    const strongLowSecondary = computeLookupScore({
+      tier: SearchLookupTier.TITLE_EXACT,
+      secondary: 0,
+    });
+    const weakHighSecondary = computeLookupScore({
+      tier: SearchLookupTier.TEXT,
+      secondary: 1e9,
+    });
+    expect(strongLowSecondary).toBeGreaterThan(weakHighSecondary);
+  });
+
+  it('within a tier a larger secondary sorts higher', () => {
+    const lo = computeLookupScore({
+      tier: SearchLookupTier.TEXT,
+      secondary: 0.1,
+    });
+    const hi = computeLookupScore({
+      tier: SearchLookupTier.TEXT,
+      secondary: 5,
+    });
+    expect(hi).toBeGreaterThan(lo);
+  });
+
+  it('treats a negative/absent secondary as 0', () => {
+    const zero = computeLookupScore({ tier: SearchLookupTier.TEXT, secondary: 0 });
+    expect(computeLookupScore({ tier: SearchLookupTier.TEXT })).toBe(zero);
+    expect(
+      computeLookupScore({ tier: SearchLookupTier.TEXT, secondary: -5 }),
+    ).toBe(zero);
+  });
+});

apps/server/src/core/search/search.service.ts

@@ -1,6 +1,9 @@
 import { Injectable } from '@nestjs/common';
 import { SearchDTO, SearchSuggestionDTO } from './dto/search.dto';
-import { SearchResponseDto } from './dto/search-response.dto';
+import {
+  SearchLookupResponseDto,
+  SearchResponseDto,
+} from './dto/search-response.dto';
 import { InjectKysely } from 'nestjs-kysely';
 import { KyselyDB } from '@docmost/db/types/kysely.types';
 import { sql } from 'kysely';
@@ -34,6 +37,53 @@ export function buildTsQuery(raw: string): string {
   return tsquery(cleaned + '*');
 }
 
+// Escape the LIKE metacharacters (`%`, `_`, `\`) in a raw user query so every
+// character — including `.`, `-`, `_`, `%`, `/` — is matched LITERALLY by a
+// `col LIKE '%' || q || '%'` predicate. Without this, a query of `%` or `_`
+// would match every row (see the #443 acceptance table). The backslash is the
+// escape char (Postgres LIKE default), so it must be escaped first.
+export function escapeLikePattern(raw: string): string {
+  return (raw ?? '')
+    .replace(/\\/g, '\\\\')
+    .replace(/%/g, '\\%')
+    .replace(/_/g, '\\_');
+}
+
+// Ranking tiers for the agent-lookup mode (#443), highest first. A hit's tier
+// is the strongest way it matched; ties inside a tier break on a secondary
+// signal (FTS rank, or first-match position). The numeric `score` returned to
+// the caller is derived from (tier, secondary) and is meaningful ONLY for
+// ordering within a single response.
+export enum SearchLookupTier {
+  // Title equals the query, case-insensitively.
+  TITLE_EXACT = 3,
+  // Query is a substring of the title.
+  TITLE_SUBSTRING = 2,
+  // Query matched in the text (substring or FTS).
+  TEXT = 1,
+}
+
+export interface RankableHit {
+  tier: SearchLookupTier;
+  // Secondary in-tier signal, higher = better (e.g. ts_rank, or a
+  // position-derived closeness score). Defaults to 0.
+  secondary?: number;
+}
+
+// Map (tier, secondary) → a 0..1 float used ONLY to sort one response.
+//
+// Formula: score = (tier + squash(secondary)) / (maxTier + 1), where
+//   squash(x) = x / (1 + x)  maps any non-negative secondary into [0, 1)
+// so a stronger tier ALWAYS outranks a weaker one regardless of the secondary
+// value, and within a tier a larger secondary sorts higher. maxTier is the top
+// enum value (TITLE_EXACT = 3), so the divisor keeps the result in (0, 1].
+export function computeLookupScore(hit: RankableHit): number {
+  const maxTier = SearchLookupTier.TITLE_EXACT;
+  const secondary = Math.max(0, hit.secondary ?? 0);
+  const squashed = secondary / (1 + secondary);
+  return (hit.tier + squashed) / (maxTier + 1);
+}
+
 @Injectable()
 export class SearchService {
   constructor(
@@ -50,12 +100,19 @@ export class SearchService {
       userId?: string;
       workspaceId: string;
     },
-  ): Promise<{ items: SearchResponseDto[] }> {
+  ): Promise<{ items: SearchResponseDto[] | SearchLookupResponseDto[] }> {
     const { query } = searchParams;
 
     if (query.length < 1) {
       return { items: [] };
     }
+
+    // Opt-in agent-lookup mode (#443). Guarded by the `substring` flag so the
+    // web-UI (which never sets it) keeps byte-identical FTS behaviour below.
+    if (searchParams.substring) {
+      return this.searchPageLookup(searchParams, opts);
+    }
+
     const searchQuery = buildTsQuery(query);
 
     let queryResults = this.db
@@ -175,6 +232,348 @@ export class SearchService {
     return { items: searchResults };
   }
 
+  /**
+   * Agent-lookup search (#443, opt-in via `SearchDTO.substring`).
+   *
+   * ADDITIVE to the FTS path: runs a substring branch (title + optionally
+   * text_content, LIKE with metacharacters escaped) MERGED with the existing
+   * FTS branch, so technical tokens that the `english` tokenizer mangles
+   * (`backup-srv.local`, `10.0.12.5`, `WB-MGE-30D86B`) are still found — even
+   * when `buildTsQuery()` returns '' for a dotted/numeric query. Results carry a
+   * location (`path`), a windowed `snippet` and a per-response `score`.
+   *
+   * The whole method is only reached when `substring: true`; the web-UI never
+   * sets it, so its behaviour is unchanged.
+   */
+  private async searchPageLookup(
+    searchParams: SearchDTO,
+    opts: { userId?: string; workspaceId: string },
+  ): Promise<{ items: SearchLookupResponseDto[] }> {
+    const rawQuery = searchParams.query.trim();
+    if (!rawQuery) {
+      return { items: [] };
+    }
+
+    const limit = Math.min(Math.max(searchParams.limit || 10, 1), 50);
+
+    // Normalize the query the same way as the FTS / suggest path: f_unaccent +
+    // lower, done in SQL. `q` is the escaped LIKE pattern body (literal chars).
+    const likeBody = escapeLikePattern(rawQuery);
+    // Compare against `LOWER(f_unaccent(col))`; unaccent+lower the needle too.
+    const needle = sql<string>`LOWER(f_unaccent(${rawQuery}))`;
+    const likePattern = sql<string>`LOWER(f_unaccent(${'%' + likeBody + '%'}))`;
+    const tsQuery = buildTsQuery(rawQuery);
+    const hasTsQuery = tsQuery.length > 0;
+
+    // --- Resolve the space scope. ---------------------------------------------
+    // Mirrors searchPage: explicit spaceId, else the authenticated user's member
+    // spaces. The share path is not exposed to this opt-in mode.
+    let spaceIds: string[] = [];
+    if (searchParams.spaceId) {
+      spaceIds = [searchParams.spaceId];
+    } else if (opts.userId) {
+      spaceIds = await this.spaceMemberRepo.getUserSpaceIds(opts.userId);
+    } else {
+      return { items: [] };
+    }
+    if (spaceIds.length === 0) {
+      return { items: [] };
+    }
+
+    // --- Optional parentPageId subtree scope (inclusive). ---------------------
+    // Reuse the same recursive-descendants pattern used for share-scope.
+    let descendantIds: string[] | null = null;
+    if (searchParams.parentPageId) {
+      const descendants = await this.pageRepo.getPageAndDescendants(
+        searchParams.parentPageId,
+        { includeContent: false },
+      );
+      descendantIds = descendants.map((p: any) => p.id);
+      if (descendantIds.length === 0) {
+        return { items: [] };
+      }
+    }
+
+    // --- Candidate query: substring (title + text) UNION FTS. -----------------
+    // We compute everything the ranker needs in SQL and pull only small columns
+    // (never the whole text_content) into Node:
+    //   - titleExact / titleSub: tier signals
+    //   - textMatchPos: 1-based position of the first text match (0 = none)
+    //   - ftsRank: ts_rank for the FTS secondary signal (0 when no tsquery)
+    //   - snippet: windowed ~500 chars around the first text match, or a leading
+    //     text window (title-only hit), or an extended ts_headline fallback.
+    const N_BEFORE = 60; // chars of context before the first match
+    const SNIPPET_LEN = 500;
+
+    let candidates = this.db
+      .selectFrom('pages')
+      .select([
+        'pages.id as id',
+        'pages.slugId as slugId',
+        'pages.title as title',
+        'pages.parentPageId as parentPageId',
+        // Tier signals.
+        sql<boolean>`LOWER(f_unaccent(coalesce(pages.title, ''))) = ${needle}`.as(
+          'titleExact',
+        ),
+        sql<boolean>`LOWER(f_unaccent(coalesce(pages.title, ''))) LIKE ${likePattern} ESCAPE '\\'`.as(
+          'titleSub',
+        ),
+        // 1-based position of the first text match (0 = no text match).
+        sql<number>`strpos(LOWER(f_unaccent(coalesce(pages.text_content, ''))), ${needle})`.as(
+          'textMatchPos',
+        ),
+        // FTS secondary signal (0 when the tsquery is empty).
+        hasTsQuery
+          ? sql<number>`ts_rank(pages.tsv, to_tsquery('english', f_unaccent(${tsQuery})))`.as(
+              'ftsRank',
+            )
+          : sql<number>`0`.as('ftsRank'),
+        // Windowed snippet, computed entirely in SQL. Priority:
+        //  1. window around the first text match;
+        //  2. otherwise (titleOnly: no snippet; else) a leading window of the
+        //     page text (title-only hit);
+        //  3. otherwise an extended ts_headline for pure-FTS hits.
+        //
+        // #443 snippet-position fix: the match position (`strpos`) is computed in
+        // the LOWER(f_unaccent(...)) space, but f_unaccent is NOT length-
+        // preserving (ß→ss, æ→ae, …→..., ½→ 1/2, full-width forms), so slicing
+        // the ORIGINAL text at that position was misaligned — a single expanding
+        // char before the match shifted the window (or ran it past end → empty).
+        // We now slice from the SAME LOWER(f_unaccent(...)) string so position
+        // and slice share one coordinate space. DELIBERATE trade-off: the snippet
+        // loses original case/diacritics — acceptable for an agent-facing snippet
+        // (position accuracy over original-glyph fidelity). The ts_headline branch
+        // matches over the ORIGINAL text itself, so it is unaffected and kept as-is.
+        searchParams.titleOnly
+          ? sql<string>`''`.as('snippet')
+          : sql<string>`
+          coalesce(
+            case
+              when strpos(LOWER(f_unaccent(coalesce(pages.text_content, ''))), ${needle}) > 0
+                then substring(
+                  LOWER(f_unaccent(coalesce(pages.text_content, '')))
+                  from greatest(1, strpos(LOWER(f_unaccent(coalesce(pages.text_content, ''))), ${needle}) - ${N_BEFORE})
+                  for ${SNIPPET_LEN}
+                )
+              when coalesce(pages.text_content, '') <> ''
+                then substring(LOWER(f_unaccent(pages.text_content)) from 1 for 300)
+              ${
+                hasTsQuery
+                  ? sql`else ts_headline('english', coalesce(pages.text_content, ''), to_tsquery('english', f_unaccent(${tsQuery})), 'MinWords=25, MaxWords=40, MaxFragments=3')`
+                  : sql``
+              }
+            end,
+            ''
+          )
+        `.as('snippet'),
+      ])
+      .where('pages.deletedAt', 'is', null)
+      .where('pages.spaceId', 'in', spaceIds);
+
+    if (descendantIds) {
+      candidates = candidates.where('pages.id', 'in', descendantIds);
+    }
+
+    // Match predicate: title substring OR (unless titleOnly) text substring OR
+    // (unless titleOnly) FTS. The substring branch runs even when the tsquery is
+    // empty — that is the dotted/numeric-token case the FTS path misses.
+    //
+    // #443 dead-index fix: these two LIKE predicates MUST match the GIN trgm
+    // index expressions EXACTLY for Postgres to use them. The indexes are on the
+    // coalesce-FREE expressions `LOWER(f_unaccent(title))` (#348's
+    // idx_pages_title_trgm) and `LOWER(f_unaccent(text_content))` (this PR's
+    // idx_pages_text_content_trgm). A `coalesce(col,'')` wrapper here would make
+    // the query expression differ from the index expression and force a Seq Scan
+    // on pages for every lookup. Dropping coalesce is SEMANTICALLY EQUIVALENT:
+    // `NULL LIKE '%q%'` is NULL (falsy), so a NULL title/text simply doesn't
+    // match — exactly as an empty string wouldn't match `%q%`.
+    candidates = candidates.where((eb) => {
+      const ors = [
+        eb(
+          sql`LOWER(f_unaccent(pages.title))`,
+          'like',
+          sql`${likePattern} ESCAPE '\\'`,
+        ),
+      ];
+      if (!searchParams.titleOnly) {
+        ors.push(
+          eb(
+            sql`LOWER(f_unaccent(pages.text_content))`,
+            'like',
+            sql`${likePattern} ESCAPE '\\'`,
+          ),
+        );
+        if (hasTsQuery) {
+          ors.push(
+            sql<boolean>`pages.tsv @@ to_tsquery('english', f_unaccent(${tsQuery}))` as any,
+          );
+        }
+      }
+      return eb.or(ors);
+    });
+
+    // Pull a generous candidate set (before permission filtering + limit).
+    // Cap it so a pathological match set cannot blow up memory; 200 >> limit
+    // (max 50) leaves ample headroom for the post-permission truncation.
+    //
+    // #443 cap-ordering fix: the 200-cap MUST be deterministic and relevance-
+    // biased. Without an ORDER BY, Postgres returns an ARBITRARY 200 rows, so on
+    // a broad match set (common word / short substring) a strong TITLE_EXACT hit
+    // could be among the dropped rows while 200 low-tier TEXT hits fill the cap.
+    // We order by the SAME SQL tier proxies the Node ranker uses — title-exact,
+    // then title-substring, then fts-rank (nulls last), then earliest text-match
+    // position — so the cap keeps the strongest candidates. The Node-side final
+    // tier sort + slice(0, limit) below still runs and stays authoritative; this
+    // ORDER BY only decides WHICH candidates survive the 200-cap.
+    // NB: a BARE integer literal in ORDER BY is read by Postgres as an ordinal
+    // column position (`ORDER BY 0` → "position 0 is not in select list"), so the
+    // no-tsquery fallback is `0::float`, not `0`.
+    const ftsRankExpr = hasTsQuery
+      ? sql`ts_rank(pages.tsv, to_tsquery('english', f_unaccent(${tsQuery})))`
+      : sql`0::float`;
+    const candidatesCapped = candidates
+      // Raw-SQL ORDER BY expressions: pass the full `<expr> <dir>` as ONE arg
+      // (the two-arg form treats a raw-SQL second arg as an ORDER BY position).
+      .orderBy(
+        sql`(LOWER(f_unaccent(coalesce(pages.title, ''))) = ${needle}) desc`,
+      )
+      .orderBy(
+        sql`(LOWER(f_unaccent(coalesce(pages.title, ''))) LIKE ${likePattern} ESCAPE '\\') desc`,
+      )
+      .orderBy(sql`${ftsRankExpr} desc nulls last`)
+      // Earlier text match first; strpos returns 0 for "no match", which would
+      // sort BEFORE a real (>=1) position under plain ASC, so push 0 to the end.
+      .orderBy(
+        sql`case when strpos(LOWER(f_unaccent(coalesce(pages.text_content, ''))), ${needle}) = 0 then 2147483647 else strpos(LOWER(f_unaccent(coalesce(pages.text_content, ''))), ${needle}) end asc`,
+      );
+
+    let rows: any[] = await candidatesCapped.limit(200).execute();
+
+    if (rows.length === 0) {
+      return { items: [] };
+    }
+
+    // --- Permissions BEFORE limit. --------------------------------------------
+    // Apply the existing page-level post-filter to the MERGED set, then rank and
+    // only THEN truncate to `limit` — never lose the permission filter.
+    if (opts.userId) {
+      const accessibleIds =
+        await this.pagePermissionRepo.filterAccessiblePageIds({
+          pageIds: rows.map((r) => r.id),
+          userId: opts.userId,
+          spaceId: searchParams.spaceId,
+          workspaceId: opts.workspaceId,
+        });
+      const accessibleSet = new Set(accessibleIds);
+      rows = rows.filter((r) => accessibleSet.has(r.id));
+    }
+
+    if (rows.length === 0) {
+      return { items: [] };
+    }
+
+    // --- Tiered ranking + dedup. ----------------------------------------------
+    // Rows are already unique by id (single pages scan), so no cross-branch
+    // dedup is needed here; the tier captures the strongest match reason.
+    const ranked = rows.map((r) => {
+      let tier: SearchLookupTier;
+      let secondary: number;
+      if (r.titleExact) {
+        tier = SearchLookupTier.TITLE_EXACT;
+        secondary = Number(r.ftsRank) || 0;
+      } else if (r.titleSub) {
+        tier = SearchLookupTier.TITLE_SUBSTRING;
+        secondary = Number(r.ftsRank) || 0;
+      } else {
+        tier = SearchLookupTier.TEXT;
+        // Prefer earlier text matches; map position → closeness in (0, 1].
+        const pos = Number(r.textMatchPos) || 0;
+        secondary =
+          pos > 0 ? 1 / (1 + (pos - 1) / 100) : Number(r.ftsRank) || 0;
+      }
+      return { row: r, tier, score: computeLookupScore({ tier, secondary }) };
+    });
+
+    ranked.sort((a, b) => b.score - a.score);
+    const top = ranked.slice(0, limit);
+
+    // --- Batch ancestor path (ONE recursive CTE, not N+1). --------------------
+    const pathById = await this.buildAncestorPaths(top.map((t) => t.row.id));
+
+    const items: SearchLookupResponseDto[] = top.map((t) => ({
+      id: t.row.id,
+      slugId: t.row.slugId,
+      title: t.row.title,
+      parentPageId: t.row.parentPageId ?? null,
+      path: pathById.get(t.row.id) ?? [],
+      snippet: (t.row.snippet ?? '')
+        .replace(/\r\n|\r|\n/g, ' ')
+        .replace(/\s+/g, ' ')
+        .trim(),
+      score: t.score,
+    }));
+
+    return { items };
+  }
+
+  /**
+   * Batch ancestor-titles helper (#443): ONE recursive CTE seeded with ALL hit
+   * ids, walking UP parentPageId. Returns a map hitId → ancestor titles ordered
+   * root → direct parent (the hit's own title is excluded). Root pages map to
+   * an empty array. Avoids the N+1 of a per-page breadcrumb call.
+   */
+  private async buildAncestorPaths(
+    hitIds: string[],
+  ): Promise<Map<string, string[]>> {
+    const result = new Map<string, string[]>();
+    if (hitIds.length === 0) return result;
+
+    // ancestry(hit_id, page_id, title, parent_page_id, depth): seed one row per
+    // hit at depth 0 (the hit itself), then walk to parents (increasing depth).
+    const rows = await this.db
+      .withRecursive('ancestry', (db) =>
+        db
+          .selectFrom('pages')
+          .select([
+            'pages.id as hitId',
+            'pages.id as pageId',
+            'pages.title as title',
+            'pages.parentPageId as parentPageId',
+            sql<number>`0`.as('depth'),
+          ])
+          .where('pages.id', 'in', hitIds)
+          .unionAll((exp) =>
+            exp
+              .selectFrom('pages as p')
+              .innerJoin('ancestry as a', 'p.id', 'a.parentPageId')
+              .select([
+                'a.hitId as hitId',
+                'p.id as pageId',
+                'p.title as title',
+                'p.parentPageId as parentPageId',
+                sql<number>`a.depth + 1`.as('depth'),
+              ]),
+          ),
+      )
+      .selectFrom('ancestry')
+      .select(['hitId', 'title', 'depth'])
+      // depth 0 is the hit itself — excluded from the path.
+      .where('depth', '>', 0)
+      .orderBy('hitId')
+      // Larger depth = closer to the space root. Ordering DESC gives
+      // root → parent once collected.
+      .orderBy('depth', 'desc')
+      .execute();
+
+    for (const r of rows as any[]) {
+      const list = result.get(r.hitId) ?? [];
+      list.push(r.title);
+      result.set(r.hitId, list);
+    }
+    return result;
+  }
+
   async searchSuggestions(
     suggestion: SearchSuggestionDTO,
     userId: string,

apps/server/src/database/migrations/20260706T120000-search-lookup-trgm.ts

@@ -0,0 +1,55 @@
+import { type Kysely, sql } from 'kysely';
+
+/**
+ * #443 — trigram indexes for the opt-in agent-lookup search mode.
+ *
+ * The lookup mode adds a substring branch that runs leading-wildcard
+ * `LOWER(f_unaccent(col)) LIKE '%q%'` predicates on pages.title and
+ * pages.text_content. A leading wildcard cannot use a b-tree index, so without a
+ * GIN trigram index each such predicate is a sequential scan.
+ *
+ *  - TITLE: the lookup-mode title predicate is `LOWER(f_unaccent(title)) LIKE
+ *    '%q%'` (coalesce-free, so it can use a functional index), which is IDENTICAL
+ *    to the one added for /search/suggest (#348). #348's perf-indexes migration
+ *    already created `idx_pages_title_trgm` on `(LOWER(f_unaccent(title)))
+ *    gin_trgm_ops`, so the title predicate is already covered — we do NOT
+ *    re-create that index here (it would be redundant).
+ *
+ *  - TEXT_CONTENT: NEW. The substring branch scans text_content when the query
+ *    is not titleOnly. text_content is the large column, so a GIN trigram index
+ *    on it is the meaningful acceleration for the lookup mode. The lookup search
+ *    is ALWAYS space-scoped (spaceId or the user's member spaces), so on small
+ *    instances a per-space sequential scan is tolerable — but the index turns the
+ *    `%q%` text predicate into a Bitmap Index Scan and removes the only
+ *    unbounded-per-space cost of the feature. We add it. The trade-off is disk +
+ *    write amplification on page edits (GIN trigram indexes are larger and slower
+ *    to update than b-trees); on the small instances this fork targets that cost
+ *    is acceptable and the read win on agent lookups is the priority.
+ *
+ * DEPLOY-TIME LOCK WARNING: plain (non-CONCURRENT) CREATE INDEX — Kysely runs
+ * each migration in a transaction, so CONCURRENTLY is impossible. The build takes
+ * a SHARE lock that BLOCKS writes on `pages` for its duration. The text_content
+ * GIN build is the slow one and can take minutes on a large tenant. For large
+ * installations, run this in a maintenance window or build the index out-of-band
+ * with CREATE INDEX CONCURRENTLY before deploying (then `IF NOT EXISTS` no-ops
+ * here). Small/typical tenants are unaffected.
+ */
+export async function up(db: Kysely<any>): Promise<void> {
+  // The title predicate is served by #348's idx_pages_title_trgm — see header.
+  // Only the text_content index is introduced here.
+
+  // text_content trigram index. Its expression is coalesce-free —
+  // `LOWER(f_unaccent(text_content))` — to EXACTLY match the coalesce-free
+  // lookup-mode text substring predicate in search.service.ts, so Postgres can
+  // use it (a `coalesce(...)` mismatch would silently fall back to a Seq Scan).
+  await sql`
+    CREATE INDEX IF NOT EXISTS idx_pages_text_content_trgm
+      ON pages USING gin ((LOWER(f_unaccent(text_content))) gin_trgm_ops)
+  `.execute(db);
+}
+
+export async function down(db: Kysely<any>): Promise<void> {
+  // Only drop the index this migration introduced. idx_pages_title_trgm is owned
+  // by the #348 perf-indexes migration, so leave it for that migration's down().
+  await sql`DROP INDEX IF EXISTS idx_pages_text_content_trgm`.execute(db);
+}

apps/server/src/integrations/import/utils/foreign-markdown.ts

@@ -238,8 +238,9 @@ function convertReferenceFootnotes(markdown: string): string {
  *
  * LINE-ANCHORED (the same shape the canonical parser uses in
  * prosemirror-markdown/page-file.ts): the block opens only on `---\n` at the
- * very start and closes only on a `\n---` line. The retired `markdownToHtml`
+ * very start and closes only on a `\n---` line. The retired editor-ext
- * strip closed on the FIRST `---` ANYWHERE (an unanchored close), so a value
+ * `markdownToHtml` front-matter strip (removed in #347) closed on the FIRST
+ * `---` ANYWHERE (an unanchored close), so a value
  * containing a triple-dash (e.g. `title: Q1 --- Q2`) truncated the front-matter
  * and leaked the rest into the body. An optional leading BOM is tolerated.
  */

apps/server/src/integrations/mcp/mcp.service.ts

@@ -34,6 +34,8 @@ import {
   isMetricsEnabled,
   observeMcpTool,
   incConnectTimeout,
+  incGetPageCacheHit,
+  incGetPageCacheMiss,
 } from '../metrics/metrics.registry';
 
 // Minimal shape of the embedded MCP HTTP handler exported by @docmost/mcp/http.
@@ -357,6 +359,10 @@ export class McpService implements OnModuleDestroy {
                       observeMcpTool(labels?.tool ?? 'other', value);
                     } else if (name === 'collab_connect_timeouts_total') {
                       incConnectTimeout();
+                    } else if (name === 'mcp_getpage_cache_hits_total') {
+                      incGetPageCacheHit();
+                    } else if (name === 'mcp_getpage_cache_misses_total') {
+                      incGetPageCacheMiss();
                     }
                   }
                 : undefined,

apps/server/src/integrations/metrics/metrics.constants.ts

@@ -25,6 +25,15 @@ export const METRIC_COLLAB_CONNECT_TIMEOUTS_TOTAL =
 export const METRIC_COLLAB_AUTH_DURATION = 'collab_auth_duration_seconds';
 export const METRIC_MCP_TOOL_DURATION = 'mcp_tool_duration_seconds';
 
+// #479 — getPage PM→Markdown conversion cache hit/miss counters. Emitted by the
+// MCP package via its dependency-neutral onMetric sink and routed onto these two
+// prom counters by the mcp.service onMetric callback; a >50% hit-rate is the
+// success signal for the getPage perf work. Same "do not rename" contract.
+export const METRIC_MCP_GETPAGE_CACHE_HITS_TOTAL =
+  'mcp_getpage_cache_hits_total';
+export const METRIC_MCP_GETPAGE_CACHE_MISSES_TOTAL =
+  'mcp_getpage_cache_misses_total';
+
 // Histogram buckets (seconds). Chosen to give useful p50/p95/p99 resolution
 // for typical web/DB latencies without exploding series cardinality.
 export const HTTP_BUCKETS = [

apps/server/src/integrations/metrics/metrics.registry.ts

@@ -24,6 +24,8 @@ import {
   METRIC_DB_QUERY_DURATION,
   METRIC_HTTP_REQUEST_DURATION,
   METRIC_MCP_TOOL_DURATION,
+  METRIC_MCP_GETPAGE_CACHE_HITS_TOTAL,
+  METRIC_MCP_GETPAGE_CACHE_MISSES_TOTAL,
   sizeBucket,
 } from './metrics.constants';
 
@@ -61,6 +63,9 @@ let connectTimeoutsCounter: Counter | null = null;
 let collabConnectHist: Histogram | null = null;
 let collabAuthHist: Histogram | null = null;
 let mcpToolHist: Histogram<'tool'> | null = null;
+// #479 — getPage conversion-cache hit/miss counters.
+let getPageCacheHitsCounter: Counter | null = null;
+let getPageCacheMissesCounter: Counter | null = null;
 
 // #402 — read-on-scrape source for collab_docs_open. The gauge is NEVER
 // inc/dec'd (that drifts under crashes/handoffs); instead its collect() callback
@@ -175,6 +180,18 @@ function init(): void {
     buckets: MCP_TOOL_BUCKETS,
     registers: [registry],
   });
+
+  getPageCacheHitsCounter = new Counter({
+    name: METRIC_MCP_GETPAGE_CACHE_HITS_TOTAL,
+    help: 'Total getPage PM→Markdown conversions served from the cache (skipped)',
+    registers: [registry],
+  });
+
+  getPageCacheMissesCounter = new Counter({
+    name: METRIC_MCP_GETPAGE_CACHE_MISSES_TOTAL,
+    help: 'Total getPage PM→Markdown conversions computed (cache misses)',
+    registers: [registry],
+  });
 }
 
 // Runs once when this module is first imported. Safe to call again (idempotent).
@@ -247,6 +264,14 @@ export function observeCollabAuth(seconds: number): void {
   collabAuthHist?.observe(seconds);
 }
 
+export function incGetPageCacheHit(): void {
+  getPageCacheHitsCounter?.inc();
+}
+
+export function incGetPageCacheMiss(): void {
+  getPageCacheMissesCounter?.inc();
+}
+
 export function observeMcpTool(tool: string, seconds: number): void {
   // `tool` MUST be a bounded, registration-derived MCP tool name (the caller
   // guarantees it comes from the registered-tool set) — never free-form input —

apps/server/src/integrations/metrics/metrics.spec.ts

@@ -10,6 +10,8 @@ import {
   incConnectTimeout,
   incDocLoad,
   incDocUnload,
+  incGetPageCacheHit,
+  incGetPageCacheMiss,
   isMetricsEnabled,
   observeCollabAuth,
   observeCollabConnect,
@@ -197,6 +199,8 @@ describe('metrics helpers are safe no-ops when METRICS_PORT is unset', () => {
       incDocLoad();
       incDocUnload();
       incConnectTimeout();
+      incGetPageCacheHit();
+      incGetPageCacheMiss();
       // Registering a source must not create the gauge or invoke the fn.
       registerDocsOpenSource(() => {
         throw new Error('docsOpenSource must NOT be called when disabled');

apps/server/test/integration/ai-chat-attach.int-spec.ts

@@ -182,6 +182,7 @@ describe('AiChatService run-stream attach [integration]', () => {
       {
         isAiChatDeferredToolsEnabled: () => false,
         isAiChatResumableStreamEnabled: () => true,
+        isAiChatFinalStepLockdownEnabled: () => false,
       } as any,
       registry,
     );
@@ -499,6 +500,7 @@ describe('AiChatService run-stream attach [integration]', () => {
       {
         isAiChatDeferredToolsEnabled: () => false,
         isAiChatResumableStreamEnabled: () => true,
+        isAiChatFinalStepLockdownEnabled: () => false,
       } as any,
       registry,
     );

apps/server/test/integration/ai-chat-stream.int-spec.ts

@@ -150,7 +150,7 @@ describe('AiChatService.stream [integration]', () => {
       {} as any, // pageAccess (idem)
       // environment (#332): keep deferred tool loading OFF for this lifecycle
       // harness so the toolset/behavior is exactly as before.
-      { isAiChatDeferredToolsEnabled: () => false } as any,
+      { isAiChatDeferredToolsEnabled: () => false, isAiChatFinalStepLockdownEnabled: () => false } as any,
     );
   }
 
@@ -378,7 +378,7 @@ describe('AiChatService.stream [integration]', () => {
         {} as any,
         {} as any,
         // #332: deferred tool loading ON — the property under test.
-        { isAiChatDeferredToolsEnabled: () => true } as any,
+        { isAiChatDeferredToolsEnabled: () => true, isAiChatFinalStepLockdownEnabled: () => false } as any,
       );
     }
 

apps/server/test/integration/search-lookup-explain.int-spec.ts

@@ -0,0 +1,123 @@
+import { randomUUID } from 'node:crypto';
+import { Kysely, sql } from 'kysely';
+import {
+  getTestDb,
+  destroyTestDb,
+  createWorkspace,
+  createSpace,
+} from './db';
+
+/**
+ * #443 dead-index guard — EXPLAIN on the REAL DB.
+ *
+ * The lookup mode's substring predicates run a leading-wildcard
+ * `LOWER(f_unaccent(col)) LIKE '%q%'`. Those are only fast when Postgres uses
+ * the GIN trigram indexes:
+ *   - idx_pages_title_trgm         on (LOWER(f_unaccent(title)))        [#348]
+ *   - idx_pages_text_content_trgm  on (LOWER(f_unaccent(text_content))) [#443]
+ *
+ * Postgres uses a functional index ONLY when the query expression matches the
+ * index expression EXACTLY. The original lookup query wrapped the columns in
+ * `coalesce(col,'')`, which differs from the coalesce-FREE index expression and
+ * silently forced a Seq Scan on pages for EVERY lookup (the MCP client always
+ * sends substring:true). This test locks that in.
+ *
+ * Discriminator: `SET enable_seqscan = off` asks the planner "CAN this predicate
+ * use the index at all?" — which is exactly what the coalesce bug breaks. With
+ * seqscan disabled:
+ *   - the coalesce-FREE (fixed) predicate plans a Bitmap Index Scan on the trgm
+ *     index (no Seq Scan on pages);
+ *   - the coalesce-WRAPPED (buggy) predicate cannot use the index and falls back
+ *     to a Seq Scan on pages even though seqscan is disabled.
+ * We assert both to prove the fix and to keep the regression from silently
+ * returning.
+ */
+describe('SearchService agent-lookup EXPLAIN — trgm index is live [integration]', () => {
+  let db: Kysely<any>;
+  let workspaceId: string;
+  let spaceId: string;
+
+  async function insertPage(title: string, textContent: string): Promise<void> {
+    const id = randomUUID();
+    await db
+      .insertInto('pages')
+      .values({
+        id,
+        slugId: `slug-${id.slice(0, 12)}`,
+        title,
+        textContent,
+        spaceId,
+        workspaceId,
+      })
+      .execute();
+  }
+
+  // Run EXPLAIN (no ANALYZE — we only inspect the chosen plan) and return the
+  // concatenated plan text.
+  async function explain(query: string): Promise<string> {
+    const rows = await sql<{ 'QUERY PLAN': string }>`EXPLAIN ${sql.raw(query)}`.execute(
+      db,
+    );
+    return (rows.rows as any[]).map((r) => r['QUERY PLAN']).join('\n');
+  }
+
+  beforeAll(async () => {
+    db = getTestDb();
+    workspaceId = (await createWorkspace(db)).id;
+    spaceId = (await createSpace(db, workspaceId)).id;
+
+    // Seed enough rows that a trigram index is a plausible plan. The content is
+    // varied so the '%needle%' pattern is selective.
+    for (let i = 0; i < 200; i++) {
+      await insertPage(
+        `seed-title-${i}`,
+        `seed body content number ${i} lorem ipsum dolor sit amet ${i}`,
+      );
+    }
+    await insertPage('backup-srv.local', 'the needle-token-xyz lives here');
+
+    // Keep the trgm indexes' stats fresh so the planner costs them correctly.
+    await sql`ANALYZE pages`.execute(db);
+  });
+
+  afterAll(async () => {
+    await destroyTestDb();
+  });
+
+  // Force the planner to answer "can the index be used?" rather than "is it
+  // cheaper than a seq scan on this size?". Restored after each test.
+  beforeEach(async () => {
+    await sql`SET enable_seqscan = off`.execute(db);
+  });
+  afterEach(async () => {
+    await sql`RESET enable_seqscan`.execute(db);
+  });
+
+  it('title predicate (coalesce-FREE, as fixed) uses idx_pages_title_trgm, not a Seq Scan', async () => {
+    const plan = await explain(
+      `SELECT id FROM pages WHERE LOWER(f_unaccent(title)) LIKE '%srv.local%'`,
+    );
+    expect(plan).toContain('idx_pages_title_trgm');
+    expect(plan).not.toMatch(/Seq Scan on pages/i);
+  });
+
+  it('text_content predicate (coalesce-FREE, as fixed) uses idx_pages_text_content_trgm, not a Seq Scan', async () => {
+    const plan = await explain(
+      `SELECT id FROM pages WHERE LOWER(f_unaccent(text_content)) LIKE '%needle-token%'`,
+    );
+    expect(plan).toContain('idx_pages_text_content_trgm');
+    expect(plan).not.toMatch(/Seq Scan on pages/i);
+  });
+
+  // Negative control: the OLD coalesce-wrapped predicate must NOT be able to use
+  // the index — even with seqscan disabled it can only Seq Scan pages. If this
+  // ever stops seq-scanning, the coalesce/index expressions have re-aligned and
+  // the guard above is no longer meaningful.
+  it('coalesce-WRAPPED text predicate (the bug) cannot use the index — falls to Seq Scan', async () => {
+    const plan = await explain(
+      `SELECT id FROM pages WHERE LOWER(f_unaccent(coalesce(text_content,''))) LIKE '%needle-token%'`,
+    );
+    expect(plan).not.toContain('idx_pages_text_content_trgm');
+    expect(plan).toMatch(/Seq Scan on pages/i);
+  });
+});

apps/server/test/integration/search-lookup.int-spec.ts

@@ -0,0 +1,462 @@
+import { randomUUID } from 'node:crypto';
+import { Kysely } from 'kysely';
+import { SearchService } from 'src/core/search/search.service';
+import { PageRepo } from '@docmost/db/repos/page/page.repo';
+import {
+  getTestDb,
+  destroyTestDb,
+  createWorkspace,
+  createSpace,
+} from './db';
+
+/**
+ * #443 — agent-lookup search mode, acceptance on the REAL DB schema.
+ *
+ * Exercises SearchService.searchPage(..., { substring: true }) against a
+ * migrated Postgres: substring matching of technical tokens the FTS tokenizer
+ * mangles (backup-srv.local, 10.0.12.5, WB-MGE-30D86B, "Теги: Docker"), the
+ * populated path + snippet, parentPageId subtree scoping, titleOnly, the empty
+ * result, LIKE-metacharacter escaping (`%`/`_` must NOT match everything), the
+ * permission post-filter applied BEFORE the limit, and the web-UI path staying
+ * on the legacy FTS shape when `substring` is absent.
+ *
+ * The tsv column is populated by the pages_tsvector_trigger on insert, so the
+ * FTS branch is exercised too.
+ */
+describe('SearchService agent-lookup mode [integration]', () => {
+  let db: Kysely<any>;
+  let service: SearchService;
+  let workspaceId: string;
+  let spaceId: string;
+
+  // Direct page insert (the shared createPage seeder omits text_content /
+  // parent_page_id, both of which this mode depends on). Returns the id.
+  async function insertPage(args: {
+    title: string;
+    textContent?: string;
+    parentPageId?: string | null;
+    spaceId?: string;
+  }): Promise<string> {
+    const id = randomUUID();
+    await db
+      .insertInto('pages')
+      .values({
+        id,
+        slugId: `slug-${id.slice(0, 12)}`,
+        title: args.title,
+        textContent: args.textContent ?? null,
+        parentPageId: args.parentPageId ?? null,
+        spaceId: args.spaceId ?? spaceId,
+        workspaceId,
+      })
+      .execute();
+    return id;
+  }
+
+  // Build a SearchService wired to the real DB + a real PageRepo (only its
+  // recursive-descendants method is used by this mode, and it needs only `db`),
+  // with lightweight stubs for the space-membership and permission repos so a
+  // test can drive scope + the permission post-filter explicitly.
+  function buildService(opts?: {
+    userSpaceIds?: string[];
+    // ids to KEEP after the permission post-filter; undefined = keep all.
+    accessibleIds?: string[];
+  }): SearchService {
+    const pageRepo = new PageRepo(db as any, null as any, null as any);
+    const spaceMemberRepo = {
+      getUserSpaceIds: async () => opts?.userSpaceIds ?? [spaceId],
+    };
+    const pagePermissionRepo = {
+      filterAccessiblePageIds: async ({ pageIds }: { pageIds: string[] }) =>
+        opts?.accessibleIds
+          ? pageIds.filter((id) => opts.accessibleIds!.includes(id))
+          : pageIds,
+    };
+    return new SearchService(
+      db as any,
+      pageRepo as any,
+      {} as any, // shareRepo — unused by the lookup path
+      spaceMemberRepo as any,
+      pagePermissionRepo as any,
+    );
+  }
+
+  beforeAll(async () => {
+    db = getTestDb();
+    workspaceId = (await createWorkspace(db)).id;
+    spaceId = (await createSpace(db, workspaceId)).id;
+    service = buildService();
+  });
+
+  afterAll(async () => {
+    await destroyTestDb();
+  });
+
+  it('finds `backup-srv.local` by the fragment `srv.local`', async () => {
+    const pageId = await insertPage({
+      title: 'backup-srv.local',
+      textContent: 'A backup server node.',
+    });
+
+    const { items } = (await service.searchPage(
+      { query: 'srv.local', spaceId, substring: true } as any,
+      { workspaceId },
+    )) as any;
+
+    expect(items.map((i: any) => i.id)).toContain(pageId);
+    const hit = items.find((i: any) => i.id === pageId);
+    expect(hit.title).toBe('backup-srv.local');
+    // slugId must never be part of the server response shape.
+    expect('slugId' in hit).toBe(true); // server carries it; MCP strips it
+  });
+
+  it('finds a page whose TEXT contains `10.0.12.5` by the fragment `10.0.12` (empty-tsquery case)', async () => {
+    const pageId = await insertPage({
+      title: 'Server inventory',
+      textContent: 'The backup box lives at IP: 10.0.12.5. Debian 12, backups.',
+    });
+
+    const { items } = (await service.searchPage(
+      { query: '10.0.12', spaceId, substring: true } as any,
+      { workspaceId },
+    )) as any;
+
+    const hit = items.find((i: any) => i.id === pageId);
+    expect(hit).toBeDefined();
+    // The windowed snippet must include the matched text.
+    expect(hit.snippet).toContain('10.0.12.5');
+  });
+
+  it('finds `WB-MGE-30D86B` (alphanumeric token with dashes) by title', async () => {
+    const pageId = await insertPage({
+      title: 'WB-MGE-30D86B',
+      textContent: 'Device page.',
+    });
+
+    const { items } = (await service.searchPage(
+      { query: 'WB-MGE-30D86B', spaceId, substring: true } as any,
+      { workspaceId },
+    )) as any;
+
+    const hit = items.find((i: any) => i.id === pageId);
+    expect(hit).toBeDefined();
+    // Exact title match → top tier (TITLE_EXACT=3) → score in [0.75, 1].
+    expect(hit.score).toBeGreaterThanOrEqual(0.75);
+    // And it is the top-ranked hit of its own result set.
+    expect(items[0].id).toBe(pageId);
+  });
+
+  it('finds every page whose text literally contains `Теги: Docker`', async () => {
+    const a = await insertPage({
+      title: 'Container host A',
+      textContent: 'Some notes.\nТеги: Docker, compose\nmore.',
+    });
+    const b = await insertPage({
+      title: 'Container host B',
+      textContent: 'Prelude.\nТеги: Docker\nepilogue.',
+    });
+    const noise = await insertPage({
+      title: 'Unrelated',
+      textContent: 'Теги: Kubernetes',
+    });
+
+    const { items } = (await service.searchPage(
+      { query: 'Теги: Docker', spaceId, substring: true, limit: 50 } as any,
+      { workspaceId },
+    )) as any;
+
+    const ids = items.map((i: any) => i.id);
+    expect(ids).toContain(a);
+    expect(ids).toContain(b);
+    expect(ids).not.toContain(noise);
+  });
+
+  it('populates a non-empty `path` for a nested hit and `[]` for a root hit', async () => {
+    const root = await insertPage({ title: 'Infrastructure' });
+    const mid = await insertPage({ title: 'Datacenter A', parentPageId: root });
+    const leaf = await insertPage({
+      title: 'unique-nested-host',
+      parentPageId: mid,
+    });
+
+    const { items } = (await service.searchPage(
+      { query: 'unique-nested-host', spaceId, substring: true } as any,
+      { workspaceId },
+    )) as any;
+
+    const hit = items.find((i: any) => i.id === leaf);
+    expect(hit.path).toEqual(['Infrastructure', 'Datacenter A']);
+
+    const rootHits = (await service.searchPage(
+      { query: 'Infrastructure', spaceId, substring: true } as any,
+      { workspaceId },
+    )) as any;
+    const rootHit = rootHits.items.find((i: any) => i.id === root);
+    expect(rootHit.path).toEqual([]);
+  });
+
+  it('scopes to a subtree with parentPageId (cutting off sibling branches)', async () => {
+    const branchA = await insertPage({ title: 'BranchA-root' });
+    const inA = await insertPage({
+      title: 'scoped-target-xyz',
+      parentPageId: branchA,
+    });
+    const branchB = await insertPage({ title: 'BranchB-root' });
+    const inB = await insertPage({
+      title: 'scoped-target-xyz',
+      parentPageId: branchB,
+    });
+
+    const { items } = (await service.searchPage(
+      {
+        query: 'scoped-target-xyz',
+        spaceId,
+        substring: true,
+        parentPageId: branchA,
+      } as any,
+      { workspaceId },
+    )) as any;
+
+    const ids = items.map((i: any) => i.id);
+    expect(ids).toContain(inA);
+    expect(ids).not.toContain(inB);
+  });
+
+  it('includes the parent page itself in the parentPageId subtree', async () => {
+    const parent = await insertPage({ title: 'self-included-parent' });
+    await insertPage({ title: 'child-of-self', parentPageId: parent });
+
+    const { items } = (await service.searchPage(
+      {
+        query: 'self-included-parent',
+        spaceId,
+        substring: true,
+        parentPageId: parent,
+      } as any,
+      { workspaceId },
+    )) as any;
+
+    expect(items.map((i: any) => i.id)).toContain(parent);
+  });
+
+  it('titleOnly does NOT match on text_content', async () => {
+    const pageId = await insertPage({
+      title: 'Plain title',
+      textContent: 'body mentions the-secret-token here',
+    });
+
+    const withText = (await service.searchPage(
+      { query: 'the-secret-token', spaceId, substring: true } as any,
+      { workspaceId },
+    )) as any;
+    expect(withText.items.map((i: any) => i.id)).toContain(pageId);
+
+    const titleOnly = (await service.searchPage(
+      {
+        query: 'the-secret-token',
+        spaceId,
+        substring: true,
+        titleOnly: true,
+      } as any,
+      { workspaceId },
+    )) as any;
+    expect(titleOnly.items.map((i: any) => i.id)).not.toContain(pageId);
+  });
+
+  // #443 Fix #1 regression: f_unaccent is NOT length-preserving, so an
+  // expanding char (ß→ss, …→...) BEFORE the match shifted the strpos position
+  // relative to the ORIGINAL text and the snippet slice ran past end → empty.
+  // The position and the slice now share the LOWER(f_unaccent(...)) space, so
+  // the window is aligned and always contains the matched (unaccented) token.
+  it('returns a populated snippet when an unaccent-EXPANDING char precedes the match', async () => {
+    // 300 × `ß` (each f_unaccent-expands to `ss`) before the needle. Under the
+    // old code strpos returned a position ~593 in the expanded space but the
+    // slice ran over the ORIGINAL (~360 char) text → empty snippet, match lost.
+    const prefix = 'ß'.repeat(300);
+    const pageId = await insertPage({
+      title: 'Expanding-unaccent page',
+      textContent: `${prefix} needle-token-xyz trailing.`,
+    });
+
+    const { items } = (await service.searchPage(
+      { query: 'needle-token-xyz', spaceId, substring: true } as any,
+      { workspaceId },
+    )) as any;
+
+    const hit = items.find((i: any) => i.id === pageId);
+    expect(hit).toBeDefined();
+    // Snippet must be non-empty AND contain the matched token (unaccented form).
+    expect(hit.snippet.length).toBeGreaterThan(0);
+    expect(hit.snippet).toContain('needle-token-xyz');
+  });
+
+  // #443 Fix #2 regression: >200 matching pages for a broad substring, with
+  // exactly ONE exact-title hit. Without an ORDER BY on the 200-cap the exact
+  // hit could be among the arbitrarily-dropped rows; the ORDER BY keeps the
+  // strongest candidates so it must survive the cap and rank at the top.
+  it('keeps an exact-title hit through the 200-cap on a >200-row match set', async () => {
+    const isoSpace = (await createSpace(db, workspaceId)).id;
+    const svc = buildService({ userSpaceIds: [isoSpace] });
+
+    // 250 low-tier TEXT hits: the shared substring `capword` appears only in the
+    // body, never the title, so each is a TEXT-tier match (weakest tier).
+    for (let i = 0; i < 250; i++) {
+      await insertPage({
+        title: `filler-page-${i}`,
+        textContent: `body contains capword here #${i}`,
+        spaceId: isoSpace,
+      });
+    }
+    // Exactly one EXACT-title hit for the same query token.
+    const exact = await insertPage({
+      title: 'capword',
+      textContent: 'unrelated body text',
+      spaceId: isoSpace,
+    });
+
+    const { items } = (await svc.searchPage(
+      { query: 'capword', spaceId: isoSpace, substring: true, limit: 10 } as any,
+      { workspaceId },
+    )) as any;
+
+    const ids = items.map((i: any) => i.id);
+    // The exact-title hit must survive the 200-cap and appear in the top `limit`.
+    expect(ids).toContain(exact);
+    // And, being TITLE_EXACT, it must be the single strongest hit.
+    expect(items[0].id).toBe(exact);
+  });
+
+  // #443 Fix #3: titleOnly matches only the title, so it must not leak the page
+  // body as the snippet (the old "first 300 chars of text_content" fallback).
+  it('titleOnly does NOT return a text-body snippet', async () => {
+    const pageId = await insertPage({
+      title: 'titleonly-snippet-page',
+      textContent: 'SECRET-BODY-CONTENT-NOT-IN-TITLE that must not leak.',
+    });
+
+    const { items } = (await service.searchPage(
+      {
+        query: 'titleonly-snippet-page',
+        spaceId,
+        substring: true,
+        titleOnly: true,
+      } as any,
+      { workspaceId },
+    )) as any;
+
+    const hit = items.find((i: any) => i.id === pageId);
+    expect(hit).toBeDefined();
+    // The body text must not appear in the snippet; titleOnly → empty snippet.
+    expect(hit.snippet).not.toContain('SECRET-BODY-CONTENT-NOT-IN-TITLE');
+    expect(hit.snippet).toBe('');
+  });
+
+  it('returns [] (not an error) for a query that matches nothing', async () => {
+    const { items } = (await service.searchPage(
+      {
+        query: 'zzz-no-such-string-anywhere-42',
+        spaceId,
+        substring: true,
+      } as any,
+      { workspaceId },
+    )) as any;
+    expect(items).toEqual([]);
+  });
+
+  it('a `%` query does NOT match everything (LIKE metacharacter escaped)', async () => {
+    // Fresh space so we can assert on total counts without cross-test noise.
+    const isoSpace = (await createSpace(db, workspaceId)).id;
+    const svc = buildService({ userSpaceIds: [isoSpace] });
+    await insertPage({ title: 'alpha', spaceId: isoSpace });
+    await insertPage({ title: 'beta', spaceId: isoSpace });
+    const literal = await insertPage({
+      title: '100%-coverage',
+      spaceId: isoSpace,
+    });
+
+    const { items } = (await svc.searchPage(
+      { query: '%', spaceId: isoSpace, substring: true, limit: 50 } as any,
+      { workspaceId },
+    )) as any;
+
+    const ids = items.map((i: any) => i.id);
+    // `%` is a literal → matches only the page that actually contains '%'.
+    expect(ids).toContain(literal);
+    expect(ids).not.toContain(
+      items.find((i: any) => i.title === 'alpha')?.id,
+    );
+    expect(items.length).toBe(1);
+  });
+
+  it('an `_` query does NOT match everything (LIKE metacharacter escaped)', async () => {
+    const isoSpace = (await createSpace(db, workspaceId)).id;
+    const svc = buildService({ userSpaceIds: [isoSpace] });
+    await insertPage({ title: 'gamma', spaceId: isoSpace });
+    const literal = await insertPage({
+      title: 'snake_case_name',
+      spaceId: isoSpace,
+    });
+
+    const { items } = (await svc.searchPage(
+      { query: '_', spaceId: isoSpace, substring: true, limit: 50 } as any,
+      { workspaceId },
+    )) as any;
+
+    const ids = items.map((i: any) => i.id);
+    expect(ids).toContain(literal);
+    expect(items.length).toBe(1);
+  });
+
+  it('applies the permission post-filter to the MERGED set BEFORE the limit', async () => {
+    const isoSpace = (await createSpace(db, workspaceId)).id;
+    const keep = await insertPage({
+      title: 'perm-visible-target',
+      spaceId: isoSpace,
+    });
+    const hidden = await insertPage({
+      title: 'perm-hidden-target',
+      spaceId: isoSpace,
+    });
+
+    // Authenticated (userId set) so the permission filter runs; only `keep` is
+    // accessible. limit 1 must NOT be able to select `hidden`.
+    const svc = buildService({
+      userSpaceIds: [isoSpace],
+      accessibleIds: [keep],
+    });
+    const { items } = (await svc.searchPage(
+      {
+        query: 'perm-',
+        spaceId: isoSpace,
+        substring: true,
+        limit: 1,
+      } as any,
+      { userId: 'user-1', workspaceId },
+    )) as any;
+
+    const ids = items.map((i: any) => i.id);
+    expect(ids).toContain(keep);
+    expect(ids).not.toContain(hidden);
+  });
+
+  it('web-UI path (no `substring` flag) keeps the legacy FTS response shape', async () => {
+    await insertPage({
+      title: 'legacy shape page',
+      textContent: 'searchable legacyword content',
+    });
+
+    const { items } = (await service.searchPage(
+      { query: 'legacyword', spaceId } as any,
+      { userId: 'user-1', workspaceId },
+    )) as any;
+
+    // Legacy hits carry rank + highlight + space, and NO path/snippet/score.
+    const hit = items[0];
+    expect(hit).toBeDefined();
+    expect('rank' in hit).toBe(true);
+    expect('highlight' in hit).toBe(true);
+    expect('path' in hit).toBe(false);
+    expect('snippet' in hit).toBe(false);
+    expect('score' in hit).toBe(false);
+  });
+});

packages/editor-ext/package.json

@@ -11,9 +11,6 @@
   "main": "dist/index.js",
   "module": "./src/index.ts",
   "types": "dist/index.d.ts",
-  "dependencies": {
-    "marked": "17.0.5"
-  },
   "devDependencies": {
     "@vitest/coverage-v8": "4.1.6",
     "vitest": "4.1.6"

packages/editor-ext/src/index.ts

@@ -18,7 +18,6 @@ export * from "./lib/excalidraw";
 export * from "./lib/embed";
 export * from "./lib/html-embed/html-embed";
 export * from "./lib/mention";
-export * from "./lib/markdown";
 export * from "./lib/search-and-replace";
 export * from "./lib/embed-provider";
 export * from "./lib/subpages";

packages/editor-ext/src/lib/footnote/footnote-canonicalize.ts

@@ -14,7 +14,8 @@ import {
  * ProseMirror JSON directly (never running the editor's plugins), so the
  * canonical footnote topology was never enforced on those writes. The consumers
  * of this editor-ext copy are: the server markdown/HTML import
- * (`markdownToHtml -> htmlToJson` in import.service / file-import-task.service),
+ * (`markdownToProseMirror` from @docmost/prosemirror-markdown in import.service /
+ * file-import-task.service),
  * `PageService` create/update (`parseProsemirrorContent` for the JSON/markdown/
  * HTML REST write paths), and the client markdown PASTE path
  * (`markdown-clipboard.ts`). (The MCP package mirrors this canonicalizer in

packages/editor-ext/src/lib/footnote/footnote-markdown.test.ts

@@ -1,131 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { htmlToMarkdown } from "../markdown/utils/turndown.utils";
-import { markdownToHtml } from "../markdown/utils/marked.utils";
-import { extractFootnoteDefinitions } from "../markdown/utils/footnote.marked";
-
-// HTML the editor-ext nodes render (sup[data-footnote-ref], section/div).
-const HTML =
-  `<p>Water<sup data-footnote-ref data-id="fn1"></sup> and clay<sup data-footnote-ref data-id="fn2"></sup>.</p>` +
-  `<section data-footnotes>` +
-  `<div data-footnote-def data-id="fn1"><p>First note.</p></div>` +
-  `<div data-footnote-def data-id="fn2"><p>Second note.</p></div>` +
-  `</section>`;
-
-describe("footnote markdown round-trip", () => {
-  it("HTML -> Markdown produces pandoc footnote syntax", () => {
-    const md = htmlToMarkdown(HTML);
-    expect(md).toContain("[^fn1]");
-    expect(md).toContain("[^fn2]");
-    expect(md).toContain("[^fn1]: First note.");
-    expect(md).toContain("[^fn2]: Second note.");
-  });
-
-  it("Markdown -> HTML rebuilds the footnote nodes' HTML", async () => {
-    const md = htmlToMarkdown(HTML);
-    const html = await markdownToHtml(md);
-    expect(html).toContain('data-footnote-ref data-id="fn1"');
-    expect(html).toContain('data-footnote-ref data-id="fn2"');
-    expect(html).toContain("data-footnotes");
-    expect(html).toContain('data-footnote-def data-id="fn1"');
-    expect(html).toContain("First note.");
-    expect(html).toContain("Second note.");
-  });
-
-  it("preserves a [^id]: line shown inside a fenced code block (not a definition)", async () => {
-    // A document that DOCUMENTS footnote syntax inside a code fence. The
-    // `[^demo]: ...` line is example text, not a real definition, and must
-    // survive the Markdown -> HTML conversion verbatim.
-    const md = [
-      "Here is how footnotes look:",
-      "",
-      "```markdown",
-      "Some text[^demo]",
-      "",
-      "[^demo]: this is the definition",
-      "```",
-      "",
-      "End of doc.",
-    ].join("\n");
-
-    const html = await markdownToHtml(md);
-    // The example definition line is kept inside the rendered code block.
-    expect(html).toContain("[^demo]: this is the definition");
-    // It did NOT get pulled out into a real footnotes section.
-    expect(html).not.toContain("data-footnotes");
-    expect(html).not.toContain("data-footnote-def");
-  });
-
-  it("extractFootnoteDefinitions keeps the FIRST duplicate definition and reuses markers", () => {
-    // Two definitions share id `d`, and the body has two `[^d]` markers. Under
-    // the import model (#166) duplicate definition ids are FIRST-WINS: only the
-    // first definition is kept; markers are NEVER rewritten, so the two `[^d]`
-    // references reuse the single footnote.
-    const md = [
-      "See here[^d] and there[^d].",
-      "",
-      "[^d]: first",
-      "[^d]: second",
-    ].join("\n");
-
-    const { body, section } = extractFootnoteDefinitions(md);
-
-    const defIds = Array.from(
-      section.matchAll(/data-footnote-def data-id="([^"]+)"/g),
-    ).map((m) => m[1]);
-    expect(defIds).toEqual(["d"]); // first-wins: one definition
-    expect(section).toContain("first");
-    expect(section).not.toContain("second"); // duplicate dropped
-
-    // Both markers stay `[^d]` (reuse) — no `d__2` minting.
-    const refIds = Array.from(body.matchAll(/\[\^([^\]\s]+)\]/g)).map(
-      (m) => m[1],
-    );
-    expect(refIds).toEqual(["d", "d"]);
-  });
-
-  it("extractFootnoteDefinitions is DETERMINISTIC and stable (same input -> same output)", () => {
-    // The output must be a pure function of the input markdown so importing the
-    // same source twice (or via the editor and the MCP mirror) is identical.
-    const md = [
-      "See[^d] one[^d] two[^d].",
-      "",
-      "[^d]: first",
-      "[^d]: second",
-      "[^d]: third",
-    ].join("\n");
-
-    const run = () => {
-      const { body, section } = extractFootnoteDefinitions(md);
-      const defIds = Array.from(
-        section.matchAll(/data-footnote-def data-id="([^"]+)"/g),
-      ).map((m) => m[1]);
-      const refIds = Array.from(body.matchAll(/\[\^([^\]\s]+)\]/g)).map(
-        (m) => m[1],
-      );
-      return { defIds, refIds };
-    };
-
-    const a = run();
-    const b = run();
-    expect(a).toEqual(b);
-    // First-wins: one kept definition `d`; all three reuse markers stay `d`.
-    expect(a.defIds).toEqual(["d"]);
-    expect(a.refIds).toEqual(["d", "d", "d"]);
-  });
-
-  it("markdownToHtml with a reused id renders ONE shared footnote def", async () => {
-    const md = [
-      "See here[^d] and there[^d].",
-      "",
-      "[^d]: first",
-      "[^d]: second",
-    ].join("\n");
-    const html = await markdownToHtml(md);
-    const defIds = Array.from(
-      html.matchAll(/data-footnote-def data-id="([^"]+)"/g),
-    ).map((m) => m[1]);
-    expect(defIds).toEqual(["d"]); // one shared definition
-    expect(html).toContain("first");
-    expect(html).not.toContain("second");
-  });
-});

packages/editor-ext/src/lib/footnote/footnote-sync.ts

@@ -103,8 +103,9 @@ interface CollisionPlan {
  * `X__2`, `X__3`, collision-bumped) so it survives as a distinct footnote — which,
  * having no matching reference, then falls under the normal orphan policy. It is
  * only ever dropped for lacking a reference, never for colliding. The IMPORT
- * paths (footnote.marked.ts / MCP extractFootnotes) instead apply first-wins +
+ * paths (@docmost/prosemirror-markdown / MCP extractFootnotes) instead apply
- * drop + warn for duplicate definitions; that divergence is intentional — import
+ * first-wins + drop + warn for duplicate definitions; that divergence is
+ * intentional — import
  * is an agent-authored artifact we sanitize, the editor is live user data we must
  * not lose.
  *

packages/editor-ext/src/lib/footnote/footnote-util.derive-id.test.ts

@@ -6,8 +6,9 @@ import { deriveFootnoteId } from "./footnote-util";
  *
  * `deriveFootnoteId` lives ONLY in editor-ext now — it is used by
  * `resolveCollisions` (re-id of a duplicate definition) and `footnotePastePlugin`
- * (re-id of a pasted colliding definition). The MCP/marked import paths no longer
+ * (re-id of a pasted colliding definition). The MCP / @docmost/prosemirror-markdown
- * derive ids (duplicate definitions there are first-wins-dropped, #166), so there
+ * import paths no longer derive ids (duplicate definitions there are
+ * first-wins-dropped, #166), so there
  * is no cross-package copy and no parity test to keep in sync. This table pins the
  * deterministic scheme so a future change to it is a conscious one.
  */

packages/editor-ext/src/lib/footnote/footnote-util.ts

@@ -63,8 +63,9 @@ export function generateFootnoteId(): string {
  * its own seen-set before requesting the next derived id.
  *
  * Used only inside editor-ext now (resolveCollisions for a re-id'd duplicate
- * DEFINITION, and footnotePastePlugin). The MCP/marked import paths no longer
+ * DEFINITION, and footnotePastePlugin). The MCP / @docmost/prosemirror-markdown
- * derive ids — duplicate definitions there are first-wins-dropped (#166) — so
+ * import paths no longer derive ids — duplicate definitions there are
+ * first-wins-dropped (#166) — so
  * there is no cross-package copy to keep in sync. The golden table in
  * footnote-util.derive-id.test.ts pins the scheme.
  */

packages/editor-ext/src/lib/image/image-markdown.test.ts

@@ -1,68 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { generateJSON } from "@tiptap/html";
-import { Document } from "@tiptap/extension-document";
-import { Paragraph } from "@tiptap/extension-paragraph";
-import { Text } from "@tiptap/extension-text";
-import { htmlToMarkdown } from "../markdown/utils/turndown.utils";
-import { markdownToHtml } from "../markdown/utils/marked.utils";
-import { TiptapImage } from "./image";
-
-// Minimal schema for parsing markdownToHtml output back to JSON (mirrors
-// image.spec.ts), so we can assert the recovered caption EXACTLY.
-const parseExtensions = [Document, Paragraph, Text, TiptapImage];
-
-// Lossless markdown round-trip for image captions (issue #221). An image WITH a
-// caption can't be expressed as `![alt](src)`, so it is emitted as a raw <img>
-// (carrying data-caption) wrapped in a block <div>, the same trick the <video>
-// rule uses. marked passes the raw HTML through, so markdownToHtml keeps the
-// data-caption, and the image extension's parseHTML restores the attribute.
-describe("image caption markdown round-trip", () => {
-  it("HTML -> Markdown emits a raw <img data-caption> for captioned images", () => {
-    const html = `<p><img src="/files/a.png" alt="cat" data-caption="A grey cat"></p>`;
-    const md = htmlToMarkdown(html);
-    expect(md).toContain("data-caption=\"A grey cat\"");
-    expect(md).toContain('src="/files/a.png"');
-    expect(md).toContain('alt="cat"');
-    // It must NOT degrade to the lossy ![]() form.
-    expect(md).not.toContain("![cat]");
-  });
-
-  it("Markdown -> HTML restores data-caption on the <img>", async () => {
-    const html = `<p><img src="/files/a.png" alt="cat" data-caption="A grey cat"></p>`;
-    const md = htmlToMarkdown(html);
-    const back = await markdownToHtml(md);
-    expect(back).toContain('data-caption="A grey cat"');
-    expect(back).toContain('src="/files/a.png"');
-  });
-
-  it("special characters in the caption survive the round-trip (escaped)", async () => {
-    // The source caption is the decoded string `Tom & "Jerry"` (both an `&` and
-    // a `"`). escapeHtmlAttr must encode `&` -> `&amp;` and `"` -> `&quot;`.
-    const html = `<p><img src="/files/a.png" data-caption='Tom &amp; &quot;Jerry&quot;'></p>`;
-    const md = htmlToMarkdown(html);
-
-    // (a) The intermediate Markdown must carry the EXACT escaped attribute. This
-    // fails if escapeHtmlAttr stopped escaping `"` (attribute break-out:
-    // data-caption="Tom & "Jerry"") or double-encoded `&` (`&amp;amp;`).
-    expect(md).toContain('data-caption="Tom &amp; &quot;Jerry&quot;"');
-
-    const back = await markdownToHtml(md);
-    expect(back).toContain("data-caption=");
-    expect(back).toContain("Jerry");
-    expect(back).toContain("Tom");
-
-    // (b) Re-parse the rendered HTML through the image extension's parseHTML and
-    // assert the recovered caption is EXACTLY the original (no corruption, loss,
-    // or double-encoding).
-    const json = generateJSON(back, parseExtensions);
-    expect(json.content?.[0]?.attrs?.caption).toBe('Tom & "Jerry"');
-  });
-
-  it("caption-less images stay a clean ![alt](src) with no raw HTML", () => {
-    const html = `<p><img src="/files/a.png" alt="cat"></p>`;
-    const md = htmlToMarkdown(html);
-    expect(md).toContain("![cat](/files/a.png)");
-    expect(md).not.toContain("data-caption");
-    expect(md).not.toContain("<img");
-  });
-});

packages/editor-ext/src/lib/markdown/html-embed-marked.spec.ts

@@ -1,105 +0,0 @@
-import { describe, expect, it } from "vitest";
-import { htmlEmbedExtension } from "./utils/html-embed.marked";
-import { markdownToHtml } from "./index";
-import { encodeHtmlEmbedSource } from "../html-embed/html-embed";
-
-// CONTRACT tests for the marked block tokenizer that rebuilds an htmlEmbed node
-// from the `<!--html-embed:BASE64-->` marker (html-embed.marked.ts), plus the
-// observable round-trip through markdownToHtml.
-//
-// These pin the REAL tokenizer behaviour the import path depends on:
-//   - the tokenizer rule is anchored (^) and only accepts the base64 alphabet
-//     [A-Za-z0-9+/=], so a marker with non-base64 chars is NOT tokenized and
-//     survives as a literal HTML comment (not silently turned into something the
-//     server's strip no longer recognizes);
-//   - start() reports the correct index of the next marker so marked invokes the
-//     tokenizer at the right offset when a marker sits mid-document / after text;
-//   - a marker with surrounding text on the SAME line is split out into its own
-//     embed div while the surrounding text becomes ordinary paragraphs.
-//
-// The contract is asserted against the actual exported extension and pipeline —
-// no behaviour is invented; the expectations were read off the real tokenizer.
-
-const SAMPLE = "<b>x</b>";
-const ENC = encodeHtmlEmbedSource(SAMPLE);
-
-describe("htmlEmbed marked tokenizer — start()", () => {
-  it("returns the index of a marker that sits mid-document", () => {
-    const src = `hello world <!--html-embed:${ENC}-->`;
-    expect(htmlEmbedExtension.start(src)).toBe(src.indexOf("<!--html-embed:"));
-  });
-
-  it("returns 0 when the marker is at the very start", () => {
-    expect(htmlEmbedExtension.start(`<!--html-embed:${ENC}-->`)).toBe(0);
-  });
-
-  it("returns -1 when there is no marker", () => {
-    expect(htmlEmbedExtension.start("no marker here")).toBe(-1);
-  });
-});
-
-describe("htmlEmbed marked tokenizer — tokenizer()", () => {
-  it("tokenizes a marker at the start of the input, capturing the base64 payload", () => {
-    const token = htmlEmbedExtension.tokenizer(`<!--html-embed:${ENC}-->`);
-    expect(token).toBeTruthy();
-    expect(token!.type).toBe("htmlEmbed");
-    expect(token!.raw).toBe(`<!--html-embed:${ENC}-->`);
-    expect(token!.encoded).toBe(ENC);
-  });
-
-  it("tokenizes an EMPTY marker (the [A-Za-z0-9+/=]* class allows zero chars)", () => {
-    const token = htmlEmbedExtension.tokenizer("<!--html-embed:-->");
-    expect(token).toBeTruthy();
-    expect(token!.encoded).toBe("");
-    expect(token!.raw).toBe("<!--html-embed:-->");
-  });
-
-  it("does NOT tokenize when text precedes the marker (rule is anchored ^)", () => {
-    // marked relies on start() to advance to the marker; the tokenizer itself
-    // only matches at offset 0, so a non-anchored call returns undefined.
-    expect(
-      htmlEmbedExtension.tokenizer(`hello <!--html-embed:${ENC}-->`),
-    ).toBeUndefined();
-  });
-
-  it("does NOT tokenize a marker containing a non-base64 char ('$')", () => {
-    expect(
-      htmlEmbedExtension.tokenizer("<!--html-embed:ab$cd-->"),
-    ).toBeUndefined();
-  });
-
-  it("does NOT tokenize a marker containing a space", () => {
-    expect(
-      htmlEmbedExtension.tokenizer("<!--html-embed:ab cd-->"),
-    ).toBeUndefined();
-  });
-
-  it("renderer emits the embed div the node's parseHTML recognizes", () => {
-    const token = htmlEmbedExtension.tokenizer(`<!--html-embed:${ENC}-->`)!;
-    const html = htmlEmbedExtension.renderer(token as any);
-    expect(html).toBe(
-      `<div data-type="htmlEmbed" data-source="${ENC}"></div>`,
-    );
-  });
-});
-
-describe("htmlEmbed marked tokenizer — markdownToHtml round-trip", () => {
-  it("splits a marker out of surrounding same-line text into its own embed div", async () => {
-    const html = await markdownToHtml(`before <!--html-embed:${ENC}--> after`);
-    // The marker became the embed div...
-    expect(html).toContain(
-      `<div data-type="htmlEmbed" data-source="${ENC}"></div>`,
-    );
-    // ...and the surrounding text survived as ordinary paragraph content.
-    expect(html).toContain("before");
-    expect(html).toContain("after");
-  });
-
-  it("leaves a marker with non-base64 chars as a literal comment (NOT an embed div)", async () => {
-    const html = await markdownToHtml("<!--html-embed:ab$cd-->");
-    // It is NOT tokenized into an embed div the server would strip...
-    expect(html).not.toContain('data-type="htmlEmbed"');
-    // ...it passes through unchanged as a literal HTML comment.
-    expect(html).toContain("<!--html-embed:ab$cd-->");
-  });
-});

packages/editor-ext/src/lib/markdown/index.ts

@@ -1,2 +0,0 @@
-export * from "./utils/marked.utils";
-export * from "./utils/turndown.utils";

packages/editor-ext/src/lib/markdown/markdown-html-embed.spec.ts

@@ -1,112 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { markdownToHtml, htmlToMarkdown } from "./index";
-import {
-  encodeHtmlEmbedSource,
-  decodeHtmlEmbedSource,
-} from "../html-embed/html-embed";
-
-// SECURITY (Variant C admin gate, import attack surface).
-//
-// The markdown import path is the only write path where an htmlEmbed reaches
-// the server purely from file bytes (no editor / collab socket). The marked
-// tokenizer in `html-embed.marked.ts` and the turndown rule in
-// `turndown.utils.ts` are what materialize the `<!--html-embed:BASE64-->`
-// marker into the `<div data-type="htmlEmbed" data-source="BASE64">` element
-// that the server then parses into an htmlEmbed node and the admin gate strips.
-//
-// If either the tokenizer regex or the turndown rule shape drifts, the marker
-// would either (a) stop becoming an htmlEmbed node (silently dropping admin
-// content) or (b) become some OTHER tag the server's `hasHtmlEmbedNode` no
-// longer recognizes (a strip bypass). These tests pin the marker <-> embed-div
-// contract that the server-side strip relies on. editor-ext had ZERO tests
-// before this file; this adds the runner + the round-trip coverage.
-
-// The server parses the embed div by matching `data-type="htmlEmbed"` and
-// decoding `data-source`; mirror that here so the assertion is exactly what the
-// real `htmlToJson` -> htmlEmbed node parse depends on (the node's parseHTML in
-// html-embed.ts uses the same selector + decodeHtmlEmbedSource).
-const EMBED_DIV_RE = /<div[^>]*\bdata-type="htmlEmbed"[^>]*>/;
-function extractEmbedSource(html: string): string | undefined {
-  const div = EMBED_DIV_RE.exec(html);
-  if (!div) return undefined;
-  const enc = /data-source="([^"]*)"/.exec(div[0]);
-  if (!enc) return undefined;
-  return decodeHtmlEmbedSource(enc[1]);
-}
-
-// Replicates the server's `hasHtmlEmbedNode` decision against the embed *div*
-// (the HTML form the server immediately converts to JSON). If this matches, the
-// server's JSON-level `hasHtmlEmbedNode` will too, because htmlToJson maps this
-// exact div to an htmlEmbed node.
-function htmlHasHtmlEmbed(html: string): boolean {
-  return EMBED_DIV_RE.test(html);
-}
-
-describe("markdown <!--html-embed--> import round-trip", () => {
-  const source = "<script>x</script>";
-
-  it("markdownToHtml turns the marker into an htmlEmbed div carrying the source", async () => {
-    const md = "<!--html-embed:" + encodeHtmlEmbedSource(source) + "-->";
-    const html = await markdownToHtml(md);
-
-    // The marker became the embed div the server recognizes as an htmlEmbed
-    // node (so the server's hasHtmlEmbedNode would match it after htmlToJson).
-    expect(htmlHasHtmlEmbed(html)).toBe(true);
-    // The decoded source is the original script, intact.
-    expect(extractEmbedSource(html)).toBe(source);
-    // The raw script is NOT inlined into the HTML — it stays base64 in the
-    // attribute (the marker itself must not be a direct injection vector).
-    expect(html).not.toContain("<script>x</script>");
-  });
-
-  it("preserves UTF-8 / special chars in the embedded source", async () => {
-    const utf8 = '<script>console.log("héllo → 世界")</script>';
-    const md = "<!--html-embed:" + encodeHtmlEmbedSource(utf8) + "-->";
-    const html = await markdownToHtml(md);
-    expect(htmlHasHtmlEmbed(html)).toBe(true);
-    expect(extractEmbedSource(html)).toBe(utf8);
-  });
-
-  it("an empty marker still produces an htmlEmbed div (empty source)", async () => {
-    const html = await markdownToHtml("<!--html-embed:-->");
-    expect(htmlHasHtmlEmbed(html)).toBe(true);
-    expect(extractEmbedSource(html)).toBe("");
-  });
-
-  it("round-trips htmlToMarkdown -> markdownToHtml preserving the embed marker", async () => {
-    const encoded = encodeHtmlEmbedSource(source);
-    // NOTE: turndown drops a *blank* (childless) element before any custom rule
-    // runs, and the htmlEmbed div is normally childless. The export pipeline
-    // therefore must give the rule a non-blank div to fire on; we add an inert
-    // text child here to exercise the real turndown htmlEmbed rule. (A blank
-    // embed div serializing to "" is asserted separately below as a documented
-    // edge so this contract drift is visible.)
-    const startHtml = `<div data-type="htmlEmbed" data-source="${encoded}">x</div>`;
-
-    // Export to markdown: the turndown rule emits the <!--html-embed:..-->
-    // marker (lossless, inert in plain markdown viewers).
-    const md = htmlToMarkdown(startHtml);
-    expect(md).toContain("<!--html-embed:" + encoded + "-->");
-
-    // Re-import: the marker round-trips back into an embed div with the same
-    // decoded source — this is the marker <-> embed-div contract the server's
-    // import strip depends on.
-    const html = await markdownToHtml(md);
-    expect(htmlHasHtmlEmbed(html)).toBe(true);
-    expect(extractEmbedSource(html)).toBe(source);
-  });
-
-  it("documents that a BLANK embed div serializes to empty markdown (turndown drops childless blocks)", () => {
-    const encoded = encodeHtmlEmbedSource(source);
-    const blank = `<div data-type="htmlEmbed" data-source="${encoded}"></div>`;
-    // This pins current behavior so a future change to the turndown rule (e.g.
-    // making it fire on blank nodes) is caught rather than silently shipping.
-    expect(htmlToMarkdown(blank)).toBe("");
-  });
-
-  it("the base64 codec itself round-trips (no '<' leaks into the attribute)", () => {
-    const encoded = encodeHtmlEmbedSource(source);
-    expect(encoded).not.toContain("<");
-    expect(decodeHtmlEmbedSource(encoded)).toBe(source);
-  });
-});

packages/editor-ext/src/lib/markdown/utils/basename.ts

@@ -1,29 +0,0 @@
-/**
- * Flexible `basename` implementation for node and the browser
- * @see https://stackoverflow.com/a/59907288/2228771
- */
-export function getBasename(path: string) {
-  // make sure the basename is not empty, if string ends with separator
-  let end = path.length - 1;
-  while (path[end] === '/' || path[end] === '\\') {
-    --end;
-  }
-
-  // support mixing of Win + Unix path separators
-  const i1 = path.lastIndexOf('/', end);
-  const i2 = path.lastIndexOf('\\', end);
-
-  let start: number;
-  if (i1 === -1) {
-    if (i2 === -1) {
-      // no separator in the whole thing
-      return path;
-    }
-    start = i2;
-  } else if (i2 === -1) {
-    start = i1;
-  } else {
-    start = Math.max(i1, i2);
-  }
-  return path.substring(start + 1, end + 1);
-}

packages/editor-ext/src/lib/markdown/utils/callout-common.marked.ts

@@ -1,33 +0,0 @@
-/**
- * Shared pieces for the two callout tokenizers — `callout.marked.ts` (the
- * `:::type` fenced form) and `github-callout.marked.ts` (the `> [!type]` GitHub
- * alert form). Both emit the SAME callout node, so the banner type dictionary
- * and the HTML renderer live here once instead of drifting apart in two files.
- * The tokenizers themselves stay separate (different syntaxes / source matching).
- */
-
-/** The four callout banner types the editor schema supports. */
-export const CALLOUT_TYPES = ['info', 'success', 'warning', 'danger'] as const;
-
-export type CalloutType = (typeof CALLOUT_TYPES)[number];
-
-/**
- * Coerce an arbitrary type name onto a supported banner type, defaulting to
- * `info` for anything unrecognized (the shared fallback both tokenizers use).
- */
-export function normalizeCalloutType(type: string): CalloutType {
-  return (CALLOUT_TYPES as readonly string[]).includes(type)
-    ? (type as CalloutType)
-    : 'info';
-}
-
-/**
- * Render a callout node to the editor's HTML shape. `body` is the already
- * markdown-parsed inner content (marked may hand back a string synchronously).
- */
-export function renderCalloutHtml(
-  type: string,
-  body: string | Promise<string>,
-): string {
-  return `<div data-type="callout" data-callout-type="${type}">${body}</div>`;
-}

packages/editor-ext/src/lib/markdown/utils/callout.marked.ts

@@ -1,37 +0,0 @@
-import { Token, marked } from 'marked';
-import { normalizeCalloutType, renderCalloutHtml } from './callout-common.marked';
-
-interface CalloutToken {
-  type: 'callout';
-  calloutType: string;
-  text: string;
-  raw: string;
-}
-
-export const calloutExtension = {
-  name: 'callout',
-  level: 'block',
-  start(src: string) {
-    return src.match(/:::/)?.index ?? -1;
-  },
-  tokenizer(src: string): CalloutToken | undefined {
-    const rule = /^:::([a-zA-Z0-9]+)\s+([\s\S]+?):::/;
-    const match = rule.exec(src);
-
-    if (match) {
-      return {
-        type: 'callout',
-        calloutType: normalizeCalloutType(match[1]),
-        raw: match[0],
-        text: match[2].trim(),
-      };
-    }
-  },
-  renderer(token: Token) {
-    const calloutToken = token as CalloutToken;
-    return renderCalloutHtml(
-      calloutToken.calloutType,
-      marked.parse(calloutToken.text),
-    );
-  },
-};

packages/editor-ext/src/lib/markdown/utils/footnote.marked.orphan.test.ts

@@ -1,72 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { extractFootnoteDefinitions } from "./footnote.marked";
-
-/** Pull the ordered list of `data-footnote-def` ids out of the rendered section. */
-function defIds(section: string): string[] {
-  return [...section.matchAll(/data-footnote-def data-id="([^"]+)"/g)].map(
-    (m) => m[1],
-  );
-}
-
-/** Pull the ordered list of `[^id]` markers that remain in the body. */
-function bodyMarkers(body: string): string[] {
-  return [...body.matchAll(/\[\^([^\]\s]+)\]/g)].map((m) => m[1]);
-}
-
-describe("extractFootnoteDefinitions: duplicate definition ids (first-wins)", () => {
-  // Body has ONE `[^d]` reference but THREE `[^d]:` definitions. Under the
-  // import model (#166) a duplicate definition id is FIRST-WINS: only the first
-  // definition is kept; the rest are DROPPED (and surfaced by analyzeFootnotes,
-  // not silently re-id'd into orphan footnotes as before). Reference markers are
-  // never rewritten, so repeated references would reuse the single footnote.
-  const md = ["See[^d].", "", "[^d]: a", "[^d]: b", "[^d]: c"].join("\n");
-
-  it("keeps only the FIRST definition for the id (first-wins)", () => {
-    const { section } = extractFootnoteDefinitions(md);
-    const ids = defIds(section);
-    expect(ids).toEqual(["d"]);
-  });
-
-  it("keeps the first definition's text and drops the duplicates", () => {
-    const { section } = extractFootnoteDefinitions(md);
-    expect(section).toContain('data-footnote-def data-id="d"><p>a</p>');
-    // No derived `d__2` / `d__3` ids are emitted anymore.
-    expect(section).not.toContain("d__2");
-    expect(section).not.toContain("d__3");
-    // The dropped duplicate texts are not in the section.
-    expect(section).not.toContain("<p>b</p>");
-    expect(section).not.toContain("<p>c</p>");
-  });
-
-  it("leaves the SINGLE body marker as [^d] (markers are never rewritten)", () => {
-    const { body } = extractFootnoteDefinitions(md);
-    expect(bodyMarkers(body)).toEqual(["d"]);
-    expect(body).toContain("See[^d].");
-    // The definition lines themselves were pulled OUT of the body.
-    expect(body).not.toContain("[^d]: a");
-    expect(body).not.toContain("[^d]: b");
-    expect(body).not.toContain("[^d]: c");
-  });
-
-  it("does not crash and produces a well-formed footnotes section", () => {
-    const { section } = extractFootnoteDefinitions(md);
-    expect(section.startsWith("<section data-footnotes>")).toBe(true);
-    expect(section.endsWith("</section>")).toBe(true);
-    // Exactly one definition div (first-wins).
-    expect([...section.matchAll(/<div data-footnote-def/g)]).toHaveLength(1);
-  });
-});
-
-describe("extractFootnoteDefinitions: reuse (repeated references, one definition)", () => {
-  // Pandoc semantics: many `[^a]` references + one `[^a]:` definition = one
-  // footnote, shared. Markers are left intact so the editor numbers them as one.
-  const md = ["A[^a] B[^a] C[^a].", "", "[^a]: shared note"].join("\n");
-
-  it("emits exactly one definition and leaves every reference marker as [^a]", () => {
-    const { section, body } = extractFootnoteDefinitions(md);
-    expect(defIds(section)).toEqual(["a"]);
-    expect(section).toContain('data-footnote-def data-id="a"><p>shared note</p>');
-    // All three reference markers stay `a` (no `a__2`/`a__3` minting).
-    expect(bodyMarkers(body)).toEqual(["a", "a", "a"]);
-  });
-});

packages/editor-ext/src/lib/markdown/utils/footnote.marked.ts

@@ -1,131 +0,0 @@
-import { marked } from "marked";
-
-/**
- * Pandoc/GFM footnote support for the marked (Markdown -> HTML) pipeline.
- *
- * Two pieces:
- *  - an INLINE tokenizer for `[^id]` references -> <sup data-footnote-ref
- *    data-id="id"> (matches the editor-ext FootnoteReference renderHTML);
- *  - a document hook (`preprocess`/`walkTokens` is awkward for collecting +
- *    removing definitions, so we use a regex preprocessing step instead) that
- *    pulls every `[^id]: text` definition line out of the body and appends a
- *    single <section data-footnotes> with one <div data-footnote-def> per
- *    definition, so the round-trip rebuilds footnotesList + footnoteDefinition.
- *
- * Every FIRST definition line is emitted — duplicate ids are first-wins (the
- * rest are dropped, and surfaced via analyzeFootnotes), and reference markers are
- * left untouched so repeated `[^a]` references reuse the one footnote (#166).
- * Orphan definitions (no matching reference) are still emitted here; the editor's
- * sync plugin reconciles the final reference/definition set (drops orphans,
- * synthesizes a single empty definition for a reference that lacks one).
- */
-
-const DEFINITION_RE = /^\[\^([^\]\s]+)\]:[ \t]*(.*)$/;
-const REFERENCE_RE = /\[\^([^\]\s]+)\]/;
-
-interface FootnoteRefToken {
-  type: "footnoteRef";
-  raw: string;
-  id: string;
-}
-
-export const footnoteReferenceExtension = {
-  name: "footnoteRef",
-  level: "inline" as const,
-  start(src: string) {
-    return src.match(/\[\^/)?.index ?? -1;
-  },
-  tokenizer(src: string): FootnoteRefToken | undefined {
-    const match = REFERENCE_RE.exec(src);
-    // Only match at the very start of the remaining inline source.
-    if (match && match.index === 0) {
-      return {
-        type: "footnoteRef",
-        raw: match[0],
-        id: match[1],
-      };
-    }
-    return undefined;
-  },
-  renderer(token: FootnoteRefToken) {
-    return `<sup data-footnote-ref data-id="${escapeAttr(token.id)}"></sup>`;
-  },
-};
-
-function escapeAttr(value: string): string {
-  return String(value).replace(/&/g, "&amp;").replace(/"/g, "&quot;");
-}
-
-/**
- * Extract `[^id]: text` definition lines from the markdown body, returning the
- * cleaned body plus a rendered <section data-footnotes> (empty string when no
- * definitions). Call this BEFORE marked.parse and append the section to the
- * resulting HTML.
- */
-export function extractFootnoteDefinitions(markdown: string): {
-  body: string;
-  section: string;
-} {
-  const lines = markdown.split("\n");
-  const bodyLines: string[] = [];
-  const definitions: Array<{ id: string; text: string }> = [];
-
-  // Track fenced-code state so a `[^id]: ...` line that merely SHOWS footnote
-  // syntax inside a ``` / ~~~ code block is left in the body verbatim and not
-  // mistaken for a real definition.
-  let fence: string | null = null;
-
-  for (const line of lines) {
-    const fenceMatch = /^(\s*)(`{3,}|~{3,})/.exec(line);
-    if (fenceMatch) {
-      const marker = fenceMatch[2][0];
-      if (fence === null) {
-        fence = marker; // opening fence
-      } else if (marker === fence) {
-        fence = null; // closing fence (matching delimiter type)
-      }
-      bodyLines.push(line);
-      continue;
-    }
-
-    const m = fence === null ? DEFINITION_RE.exec(line) : null;
-    if (m) {
-      definitions.push({ id: m[1], text: m[2] });
-    } else {
-      bodyLines.push(line);
-    }
-  }
-
-  if (definitions.length === 0) {
-    return { body: markdown, section: "" };
-  }
-
-  // Duplicate definition ids (e.g. `[^d]: first` / `[^d]: second`): FIRST WINS,
-  // the rest are DROPPED. Reference markers are left UNTOUCHED so repeated `[^a]`
-  // references reuse the single footnote (Pandoc semantics, #166). This differs
-  // from the live editor's never-lose policy (resolveCollisions re-ids a
-  // duplicate definition into an orphan) on purpose: an import is an
-  // agent-authored artifact we sanitize, and the dropped duplicate is surfaced
-  // to the caller via analyzeFootnotes' `duplicateDefinitions` warning instead.
-  const firstById = new Map<string, string>(); // id -> first definition text
-  for (const def of definitions) {
-    if (!firstById.has(def.id)) firstById.set(def.id, def.text);
-  }
-
-  const defsHtml = [...firstById.entries()]
-    .map(([id, text]) => {
-      // Render the definition text as inline markdown so emphasis/links inside
-      // a footnote survive the round-trip; wrap in a paragraph (the node's
-      // content is paragraph+).
-      const inner = marked.parseInline(text || "");
-      return `<div data-footnote-def data-id="${escapeAttr(
-        id,
-      )}"><p>${inner}</p></div>`;
-    })
-    .join("");
-
-  return {
-    body: bodyLines.join("\n"),
-    section: `<section data-footnotes>${defsHtml}</section>`,
-  };
-}

packages/editor-ext/src/lib/markdown/utils/github-callout.marked.test.ts

@@ -1,54 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { markdownToHtml } from "./marked.utils";
-
-/**
- * Regression for issue #192: pasting a GitHub-style `> [!type]` alert produced a
- * literal `<blockquote>` containing `[!info]` instead of a callout node, because
- * only the `:::type` form was tokenized. The editor paste path runs the same
- * `markdownToHtml`, so these assertions pin the conversion at the source.
- */
-function html(md: string): string {
-  const out = markdownToHtml(md);
-  if (typeof out !== "string") throw new Error("expected sync string output");
-  return out;
-}
-
-describe("markdownToHtml: GitHub `> [!type]` callouts", () => {
-  it("converts `> [!info]` to a callout node, not a literal blockquote", () => {
-    const out = html("> [!info]\n> Callout body text here");
-    expect(out).toContain('data-type="callout"');
-    expect(out).toContain('data-callout-type="info"');
-    expect(out).toContain("Callout body text here");
-    expect(out).not.toContain("[!info]");
-    expect(out).not.toContain("<blockquote");
-  });
-
-  it("maps GitHub alert aliases onto the supported banner types", () => {
-    expect(html("> [!NOTE]\n> x")).toContain('data-callout-type="info"');
-    expect(html("> [!TIP]\n> x")).toContain('data-callout-type="success"');
-    expect(html("> [!WARNING]\n> x")).toContain('data-callout-type="warning"');
-    expect(html("> [!CAUTION]\n> x")).toContain('data-callout-type="danger"');
-  });
-
-  it("accepts the editor's own type names directly", () => {
-    expect(html("> [!success]\n> x")).toContain('data-callout-type="success"');
-    expect(html("> [!danger]\n> x")).toContain('data-callout-type="danger"');
-  });
-
-  it("falls back to info for an unknown type", () => {
-    expect(html("> [!bogus]\n> x")).toContain('data-callout-type="info"');
-  });
-
-  it("preserves multi-line callout bodies", () => {
-    const out = html("> [!warning]\n> line one\n> line two");
-    expect(out).toContain('data-callout-type="warning"');
-    expect(out).toContain("line one");
-    expect(out).toContain("line two");
-  });
-
-  it("still converts the `:::type` form", () => {
-    const out = html(":::info\nbody\n:::");
-    expect(out).toContain('data-type="callout"');
-    expect(out).toContain('data-callout-type="info"');
-  });
-});

packages/editor-ext/src/lib/markdown/utils/github-callout.marked.ts

@@ -1,81 +0,0 @@
-import { Token, marked } from 'marked';
-import { renderCalloutHtml } from './callout-common.marked';
-
-interface GithubCalloutToken {
-  type: 'githubCallout';
-  calloutType: string;
-  text: string;
-  raw: string;
-}
-
-/**
- * Map GitHub "alert" blockquote markers (`> [!NOTE]`, `> [!WARNING]`, …) onto
- * the four callout banner types the editor schema supports. The editor's own
- * type names (`info`/`success`/`warning`/`danger`) are also accepted directly,
- * because users paste both forms. Anything unrecognized falls back to `info`,
- * matching the `:::type` callout tokenizer.
- */
-const GITHUB_ALERT_TYPE_MAP: Record<string, string> = {
-  note: 'info',
-  tip: 'success',
-  important: 'info',
-  warning: 'warning',
-  caution: 'danger',
-  info: 'info',
-  success: 'success',
-  danger: 'danger',
-};
-
-/**
- * Tokenizer for GitHub-flavored alert callouts written as a blockquote whose
- * first line is `[!type]`:
- *
- *   > [!info]
- *   > body line one
- *   > body line two
- *
- * Without this, the default blockquote tokenizer wins and the marker renders as
- * a literal `[!info]` inside a `<blockquote>`. The editor's paste path runs the
- * same `markdownToHtml`, so registering this here also fixes pasting the syntax
- * into the editor (issue #192), not just markdown import.
- */
-export const githubCalloutExtension = {
-  name: 'githubCallout',
-  level: 'block' as const,
-  start(src: string) {
-    return src.match(/^ {0,3}>[ \t]*\[!/m)?.index ?? -1;
-  },
-  tokenizer(src: string): GithubCalloutToken | undefined {
-    const rule =
-      /^ {0,3}>[ \t]*\[!([a-zA-Z]+)\][^\n]*(?:\n {0,3}>[^\n]*)*(?:\n|$)/;
-    const match = rule.exec(src);
-    if (!match) return undefined;
-
-    const rawType = match[1].toLowerCase();
-    const calloutType = GITHUB_ALERT_TYPE_MAP[rawType] ?? 'info';
-
-    const text = match[0]
-      .replace(/\n+$/, '')
-      .split('\n')
-      // Strip the blockquote marker (`>` + optional space) from every line.
-      .map((line) => line.replace(/^ {0,3}>[ \t]?/, ''))
-      // Drop the `[!type]` marker that opens the first line.
-      .map((line, i) => (i === 0 ? line.replace(/^\[![a-zA-Z]+\][ \t]*/, '') : line))
-      .join('\n')
-      .trim();
-
-    return {
-      type: 'githubCallout',
-      calloutType,
-      raw: match[0],
-      text,
-    };
-  },
-  renderer(token: Token) {
-    const calloutToken = token as GithubCalloutToken;
-    return renderCalloutHtml(
-      calloutToken.calloutType,
-      marked.parse(calloutToken.text),
-    );
-  },
-};

packages/editor-ext/src/lib/markdown/utils/html-embed.marked.ts

@@ -1,41 +0,0 @@
-import { Token } from "marked";
-
-interface HtmlEmbedToken {
-  type: "htmlEmbed";
-  raw: string;
-  encoded: string;
-}
-
-/**
- * Marked extension that rebuilds an `htmlEmbed` node from the HTML comment
- * marker produced by the turndown rule (`<!--html-embed:<base64>-->`).
- *
- * It emits the same marker div the node's `parseHTML` recognizes, so the
- * pipeline MD -> HTML -> ProseMirror JSON restores the node (and its
- * base64 `data-source`) exactly. We do NOT expand the raw markup here; the
- * source stays base64-encoded in the attribute and is only executed by the
- * client NodeView.
- */
-export const htmlEmbedExtension = {
-  name: "htmlEmbed",
-  level: "block" as const,
-  start(src: string) {
-    return src.indexOf("<!--html-embed:");
-  },
-  tokenizer(src: string): HtmlEmbedToken | undefined {
-    const rule = /^<!--html-embed:([A-Za-z0-9+/=]*)-->/;
-    const match = rule.exec(src);
-
-    if (match) {
-      return {
-        type: "htmlEmbed",
-        raw: match[0],
-        encoded: match[1] ?? "",
-      };
-    }
-  },
-  renderer(token: Token) {
-    const htmlEmbedToken = token as HtmlEmbedToken;
-    return `<div data-type="htmlEmbed" data-source="${htmlEmbedToken.encoded}"></div>`;
-  },
-};

packages/editor-ext/src/lib/markdown/utils/marked.utils.ts

@@ -1,76 +0,0 @@
-import { marked } from "marked";
-import { calloutExtension } from "./callout.marked";
-import { githubCalloutExtension } from "./github-callout.marked";
-import { mathBlockExtension } from "./math-block.marked";
-import { mathInlineExtension } from "./math-inline.marked";
-import {
-  footnoteReferenceExtension,
-  extractFootnoteDefinitions,
-} from "./footnote.marked";
-import { htmlEmbedExtension } from "./html-embed.marked";
-
-marked.use({
-  renderer: {
-    list({ ordered, start, items }) {
-      let body = "";
-      for (const item of items) {
-        body += this.listitem(item);
-      }
-
-      if (ordered) {
-        const startAttr = start !== 1 ? ` start="${start}"` : "";
-        return `<ol${startAttr}>\n${body}</ol>\n`;
-      }
-
-      const isTaskList = items.some((item) => item.task);
-      const dataType = isTaskList ? ' data-type="taskList"' : "";
-      return `<ul${dataType}>\n${body}</ul>\n`;
-    },
-    listitem({ tokens, task: isTask, checked: isChecked }) {
-      const text = this.parser.parse(tokens);
-      if (!isTask) {
-        return `<li>${text}</li>\n`;
-      }
-      const checkedAttr = isChecked
-        ? 'data-checked="true"'
-        : 'data-checked="false"';
-      return `<li data-type="taskItem" ${checkedAttr}>${text}</li>\n`;
-    },
-  },
-});
-
-marked.use({
-  extensions: [
-    calloutExtension,
-    githubCalloutExtension,
-    mathBlockExtension,
-    mathInlineExtension,
-    footnoteReferenceExtension,
-    htmlEmbedExtension,
-  ],
-});
-
-marked.setOptions({ breaks: true });
-
-export function markdownToHtml(
-  markdownInput: string,
-): string | Promise<string> {
-  const YAML_FONT_MATTER_REGEX = /^\s*---[\s\S]*?---\s*/;
-
-  const markdown = markdownInput
-    .replace(YAML_FONT_MATTER_REGEX, "")
-    .trimStart();
-
-  // Pull `[^id]: ...` definition lines out of the body, render the body, then
-  // append a single <section data-footnotes> so the round-trip rebuilds the
-  // footnotesList + footnoteDefinition nodes.
-  const { body, section } = extractFootnoteDefinitions(markdown);
-
-  const parsed = marked.parse(body);
-  if (!section) return parsed;
-
-  if (typeof parsed === "string") {
-    return parsed + section;
-  }
-  return parsed.then((html) => html + section);
-}

packages/editor-ext/src/lib/markdown/utils/math-block.marked.ts

@@ -1,37 +0,0 @@
-import { Token, marked } from 'marked';
-
-interface MathBlockToken {
-  type: 'mathBlock';
-  text: string;
-  raw: string;
-}
-
-export const mathBlockExtension = {
-  name: 'mathBlock',
-  level: 'block',
-  start(src: string) {
-    return src.match(/\$\$/)?.index ?? -1;
-  },
-  tokenizer(src: string): MathBlockToken | undefined {
-    const rule = /^\$\$(?!(\$))([\s\S]+?)\$\$/;
-    const match = rule.exec(src);
-
-    if (match) {
-      return {
-        type: 'mathBlock',
-        raw: match[0],
-        text: match[2]?.trim(),
-      };
-    }
-  },
-  renderer(token: Token) {
-    const mathBlockToken = token as MathBlockToken;
-    // parse to prevent escaping slashes
-    const latex = marked
-      .parse(mathBlockToken.text)
-      .toString()
-      .replace(/<(\/)?p>/g, '');
-
-    return `<div data-type="${mathBlockToken.type}" data-katex="true">${latex}</div>`;
-  },
-};

packages/editor-ext/src/lib/markdown/utils/math-inline.marked.falsepositive.test.ts

@@ -1,50 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { markdownToHtml } from "./marked.utils";
-
-/**
- * Data-integrity regression (issue #204, Phase 2): plain prose that mentions
- * prices like `$5 and $6` must NOT be misread as inline math. The inline-math
- * tokenizer mutates a global `marked` singleton at import time
- * (`marked.utils.ts`), so math behaviour can only be exercised safely through
- * the public `markdownToHtml`; importing the tokenizer in isolation would give
- * a different, non-representative result. These assertions therefore drive the
- * real conversion path.
- */
-function html(md: string): string {
-  const out = markdownToHtml(md);
-  if (typeof out !== "string") throw new Error("expected sync string output");
-  return out;
-}
-
-const MATH_MARKERS = ['data-type="mathInline"', 'data-katex="true"'];
-
-function hasInlineMath(out: string): boolean {
-  return MATH_MARKERS.some((m) => out.includes(m));
-}
-
-describe("markdownToHtml: inline-math false positives", () => {
-  it("does not treat prices `$5 and $6` as inline math", () => {
-    const out = html("It costs $5 and $6 today.");
-    expect(hasInlineMath(out)).toBe(false);
-    // The text survives verbatim (no katex span swallowing it).
-    expect(out).toContain("$5 and $6");
-  });
-
-  it("does not treat a single trailing price `$5` as inline math", () => {
-    const out = html("Lunch was $5.");
-    expect(hasInlineMath(out)).toBe(false);
-    expect(out).toContain("$5");
-  });
-
-  it("does not treat `$5, $6, $7` (multiple prices) as inline math", () => {
-    const out = html("Choose $5, $6, $7 plans.");
-    expect(hasInlineMath(out)).toBe(false);
-  });
-
-  it("STILL converts a genuine inline-math expression `$x + y$`", () => {
-    // Guard the positive path so the false-positive guard above can't be
-    // satisfied by simply disabling math entirely.
-    const out = html("The sum $x + y$ is shown.");
-    expect(hasInlineMath(out)).toBe(true);
-  });
-});

packages/editor-ext/src/lib/markdown/utils/math-inline.marked.ts

@@ -1,55 +0,0 @@
-import { Token, marked } from 'marked';
-
-interface MathInlineToken {
-  type: 'mathInline';
-  text: string;
-  raw: string;
-}
-
-const inlineMathRegex = /^\$(?!\s)(.+?)(?<!\s)\$(?!\d)/;
-
-export const mathInlineExtension = {
-  name: 'mathInline',
-  level: 'inline',
-  start(src: string) {
-    let index: number;
-    let indexSrc = src;
-
-    while (indexSrc) {
-      index = indexSrc.indexOf('$');
-      if (index === -1) {
-        return;
-      }
-      const f = index === 0 || indexSrc.charAt(index - 1) === ' ';
-      if (f) {
-        const possibleKatex = indexSrc.substring(index);
-        if (possibleKatex.match(inlineMathRegex)) {
-          return index;
-        }
-      }
-
-      indexSrc = indexSrc.substring(index + 1).replace(/^\$+/, '');
-    }
-  },
-  tokenizer(src: string): MathInlineToken | undefined {
-    const match = inlineMathRegex.exec(src);
-
-    if (match) {
-      return {
-        type: 'mathInline',
-        raw: match[0],
-        text: match[1]?.trim(),
-      };
-    }
-  },
-  renderer(token: Token) {
-    const mathInlineToken = token as MathInlineToken;
-    // parse to prevent escaping slashes
-    const latex = marked
-      .parse(mathInlineToken.text)
-      .toString()
-      .replace(/<(\/)?p>/g, '');
-
-    return `<span data-type="${mathInlineToken.type}" data-katex="true">${latex}</span>`;
-  },
-};

packages/editor-ext/src/lib/markdown/utils/spoiler.marked.test.ts

@@ -1,128 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { getSchema } from "@tiptap/core";
-import { generateHTML, generateJSON } from "@tiptap/html";
-import { Document } from "@tiptap/extension-document";
-import { Paragraph } from "@tiptap/extension-paragraph";
-import { Text } from "@tiptap/extension-text";
-import { Bold } from "@tiptap/extension-bold";
-import { htmlToMarkdown } from "./turndown.utils";
-import { markdownToHtml } from "./marked.utils";
-import { Spoiler } from "../../spoiler/spoiler";
-
-// The spoiler mark has no native Markdown syntax, so it is preserved losslessly
-// as raw inline HTML (`<span data-spoiler="true">…</span>`), the same approach
-// htmlEmbed uses. This test drives the full editor round-trip:
-//   JSON -> HTML -> Markdown -> HTML -> JSON
-// and asserts the `spoiler` mark survives end to end. We use the same
-// getSchema + @tiptap/html generateHTML/generateJSON utilities the other
-// editor-ext schema tests use.
-
-const extensions = [Document, Paragraph, Text, Bold, Spoiler];
-
-function html(md: string): string {
-  const out = markdownToHtml(md);
-  if (typeof out !== "string") throw new Error("expected sync string output");
-  return out;
-}
-
-// Count text nodes carrying a `spoiler` mark anywhere in a ProseMirror JSON doc.
-function countSpoilerMarks(doc: any): number {
-  let count = 0;
-  const walk = (node: any) => {
-    if (!node || typeof node !== "object") return;
-    if (Array.isArray(node.marks)) {
-      for (const mark of node.marks) {
-        if (mark?.type === "spoiler") count++;
-      }
-    }
-    if (Array.isArray(node.content)) node.content.forEach(walk);
-  };
-  walk(doc);
-  return count;
-}
-
-describe("Spoiler mark schema", () => {
-  it("registers the spoiler mark in the schema", () => {
-    const schema = getSchema(extensions);
-    expect(schema.marks.spoiler).toBeTruthy();
-  });
-
-  it("recovers the spoiler mark from span[data-spoiler] (HTML -> JSON)", () => {
-    const json = generateJSON(
-      '<p>before <span data-spoiler="true">hidden</span> after</p>',
-      extensions,
-    );
-    expect(countSpoilerMarks(json)).toBe(1);
-  });
-
-  it("emits data-spoiler + class on render (JSON -> HTML)", () => {
-    const doc = {
-      type: "doc",
-      content: [
-        {
-          type: "paragraph",
-          content: [
-            {
-              type: "text",
-              text: "hidden",
-              marks: [{ type: "spoiler" }],
-            },
-          ],
-        },
-      ],
-    };
-    const out = generateHTML(doc, extensions);
-    expect(out).toContain('data-spoiler="true"');
-    expect(out).toContain('class="spoiler"');
-  });
-});
-
-describe("Spoiler Markdown round-trip is lossless", () => {
-  const docWith = (textNode: any) => ({
-    type: "doc",
-    content: [
-      {
-        type: "paragraph",
-        content: [{ type: "text", text: "before " }, textNode, { type: "text", text: " after" }],
-      },
-    ],
-  });
-
-  it("preserves the spoiler mark through JSON -> MD -> HTML -> JSON", () => {
-    const startDoc = docWith({
-      type: "text",
-      text: "hidden",
-      marks: [{ type: "spoiler" }],
-    });
-
-    // JSON -> HTML
-    const html1 = generateHTML(startDoc, extensions);
-    expect(html1).toContain('data-spoiler="true"');
-
-    // HTML -> Markdown (raw inline HTML, lossless)
-    const md = htmlToMarkdown(html1);
-    expect(md).toContain('<span data-spoiler="true">hidden</span>');
-
-    // MD -> HTML -> JSON (mark restored via parseHTML)
-    const endJson = generateJSON(html(md), extensions);
-    expect(countSpoilerMarks(endJson)).toBe(1);
-    // The visible text survives.
-    expect(JSON.stringify(endJson)).toContain("hidden");
-  });
-
-  it("keeps the spoiler intact when it intersects a bold mark", () => {
-    const startDoc = docWith({
-      type: "text",
-      text: "secret",
-      marks: [{ type: "bold" }, { type: "spoiler" }],
-    });
-
-    const md = htmlToMarkdown(generateHTML(startDoc, extensions));
-    expect(md).toContain("data-spoiler=\"true\"");
-
-    const endJson = generateJSON(html(md), extensions);
-    expect(countSpoilerMarks(endJson)).toBe(1);
-    // Bold survives alongside the spoiler.
-    expect(JSON.stringify(endJson)).toContain('"bold"');
-  });
-});

packages/editor-ext/src/lib/markdown/utils/turndown.d.ts

@@ -1,12 +0,0 @@
-// Map @joplin/turndown types to @types/turndown
-declare module "@joplin/turndown" {
-  import TurndownService from "turndown";
-  export = TurndownService;
-}
-
-declare module "@joplin/turndown-plugin-gfm" {
-  import TurndownService from "turndown";
-  export const tables: TurndownService.Plugin;
-  export const strikethrough: TurndownService.Plugin;
-  export const highlightedCodeBlock: TurndownService.Plugin;
-}

packages/editor-ext/src/lib/markdown/utils/turndown.dataloss.test.ts

@@ -1,147 +0,0 @@
-import { describe, it, expect } from "vitest";
-import { htmlToMarkdown } from "./turndown.utils";
-import { markdownToHtml } from "./marked.utils";
-
-/**
- * #206 mdrt-2 — Markdown export must never SILENTLY drop a block. (FIXED)
- *
- * `htmlToMarkdown` (turndown) historically only registered rules for a fixed
- * set of custom nodes (callout, taskItem, details, math, iframe, htmlEmbed,
- * image, video, footnote). Any other custom node — `transclusionReference`,
- * `pageBreak`, `mention`, `status` — fell through to turndown's default
- * handling: an empty wrapper is "blank" and removed, so the block disappeared
- * from the exported Markdown with no trace, and `mention`/`status` collapsed to
- * bare text, losing their identity (data-id / data-color). The invariant
- * "never silently lose a block" was broken.
- *
- * The fix adds lossless turndown rules that re-emit each of these nodes as raw
- * HTML carrying every `data-*` attribute. Plain-Markdown viewers ignore the
- * inert tag; the import path round-trips it (`markdownToHtml` passes the raw
- * HTML through and each node's `parseHTML` rebuilds the ProseMirror node). These
- * tests assert the surviving contract (the block is preserved AND its identity
- * round-trips back through import).
- */
-describe("htmlToMarkdown — custom nodes are preserved losslessly (#206 mdrt-2)", () => {
-  const wrap = (inner: string) => `<p>before</p>${inner}<p>after</p>`;
-
-  it("preserves a pageBreak block on Markdown export", () => {
-    const md = htmlToMarkdown(
-      wrap('<div data-type="pageBreak" class="page-break"></div>'),
-    );
-    expect(md).toContain("before");
-    expect(md).toContain("after");
-    // The break survives as an inert raw-HTML tag, not silently dropped.
-    expect(md).toMatch(/data-type="pageBreak"/);
-    expect(md).toMatch(/page-?break/i);
-  });
-
-  it("preserves a transclusionReference's identity on Markdown export", () => {
-    const md = htmlToMarkdown(
-      wrap('<div data-type="transclusionReference" data-id="abc"></div>'),
-    );
-    expect(md).toContain("before");
-    expect(md).toContain("after");
-    // The data-id (the only thing that gives the reference identity) survives.
-    expect(md).toContain("abc");
-    expect(md).toMatch(/data-type="transclusionReference"/);
-  });
-
-  it("preserves a mention's data-id (stable identity) on Markdown export", () => {
-    const md = htmlToMarkdown(
-      '<p>hi <span data-type="mention" data-id="u1" data-label="Bob">@Bob</span> there</p>',
-    );
-    // The mention keeps its stable identity (data-id), not just the text.
-    expect(md).toContain("u1");
-    expect(md).toContain("Bob");
-    expect(md).toMatch(/data-type="mention"/);
-  });
-
-  it("preserves a status chip's color on Markdown export", () => {
-    const md = htmlToMarkdown(
-      '<p>s <span data-type="status" data-color="green">Done</span></p>',
-    );
-    // The chip's color (its identity) survives, not just the visible text.
-    expect(md).toContain("green");
-    expect(md).toContain("Done");
-    expect(md).toMatch(/data-type="status"/);
-  });
-
-  // The export form is only lossless if the import path can rebuild it. These
-  // assert the full MD -> HTML round-trip restores the node + its attributes,
-  // which is the marker <-> node contract each `parseHTML` relies on.
-  describe("import round-trip (markdownToHtml restores the node)", () => {
-    it("round-trips a pageBreak through export + import", async () => {
-      const md = htmlToMarkdown(
-        wrap('<div data-type="pageBreak" class="page-break"></div>'),
-      );
-      const html = await markdownToHtml(md);
-      expect(html).toMatch(/<div[^>]*data-type="pageBreak"[^>]*>/);
-      expect(html).toContain("before");
-      expect(html).toContain("after");
-    });
-
-    it("round-trips a transclusionReference (keeps data-id)", async () => {
-      const md = htmlToMarkdown(
-        wrap('<div data-type="transclusionReference" data-id="abc"></div>'),
-      );
-      const html = await markdownToHtml(md);
-      expect(html).toMatch(/<div[^>]*data-type="transclusionReference"[^>]*>/);
-      expect(html).toContain("abc");
-    });
-
-    it("round-trips a mention (keeps data-id + data-label)", async () => {
-      const md = htmlToMarkdown(
-        '<p>hi <span data-type="mention" data-id="u1" data-label="Bob">@Bob</span> there</p>',
-      );
-      const html = await markdownToHtml(md);
-      expect(html).toMatch(/<span[^>]*data-type="mention"[^>]*>/);
-      expect(html).toContain("u1");
-      expect(html).toContain("Bob");
-    });
-
-    it("round-trips a status chip (keeps data-color)", async () => {
-      const md = htmlToMarkdown(
-        '<p>s <span data-type="status" data-color="green">Done</span></p>',
-      );
-      const html = await markdownToHtml(md);
-      expect(html).toMatch(/<span[^>]*data-type="status"[^>]*>/);
-      expect(html).toContain("green");
-    });
-
-    // HTML special chars in an attribute value or in a node's text must be
-    // ESCAPED when re-emitted as raw HTML, otherwise the exported tag is
-    // malformed and `markdownToHtml`'s parser cannot restore the original value
-    // (the same silent data loss this PR fixes). Dropping `<`/`>` escaping is the
-    // dangerous regression: a stray `<` or `>` corrupts the tag (or injects new
-    // markup), so the test data carries ALL of `&`, `"`, `<`, `>` in BOTH the
-    // data-label attribute and the visible text. That fully exercises
-    // escapeHtmlAttr's `&,",<,>` branches and escapeHtmlText's `&,<,>` branches
-    // (escapeHtmlText leaves `"` literal); the alphanumeric-only cases above hit
-    // none of them.
-    it("escapes HTML special chars (& \" < >) in attrs + text and round-trips them", async () => {
-      const md = htmlToMarkdown(
-        `<p>hi <span data-type="mention" data-id="u1" data-label="A &amp; &lt;B&gt; &quot;C&quot;">@A &amp; &lt;B&gt; "C"</span> there</p>`,
-      );
-
-      // (a) The exported Markdown carries a WELL-FORMED, correctly-escaped tag:
-      // the attribute escapes `&`, `<`, `>` AND `"`; the text escapes `&`, `<`,
-      // `>` (a `"` inside text content is legal, so it stays literal).
-      expect(md).toContain('data-label="A &amp; &lt;B&gt; &quot;C&quot;"');
-      expect(md).toContain('>@A &amp; &lt;B&gt; "C"</span>');
-      // And explicitly NOT the raw, tag-corrupting forms: a literal `<B>` (would
-      // mean `<`/`>` escaping was dropped in either the attr or the text)...
-      expect(md).not.toContain("<B>");
-      // ...nor the malformed attribute that an unescaped `"` would produce.
-      expect(md).not.toContain('data-label="A &amp; &lt;B&gt; "C""');
-
-      // (b) Import restores the ORIGINAL (unescaped) values, attribute and text.
-      const html = await markdownToHtml(md);
-      const dom = new DOMParser().parseFromString(html as string, "text/html");
-      const span = dom.querySelector('span[data-type="mention"]');
-      expect(span).not.toBeNull();
-      expect(span!.getAttribute("data-id")).toBe("u1");
-      expect(span!.getAttribute("data-label")).toBe('A & <B> "C"');
-      expect(span!.textContent).toBe('@A & <B> "C"');
-    });
-  });
-});

packages/editor-ext/src/lib/markdown/utils/turndown.utils.ts

@@ -1,488 +0,0 @@
-import * as _TurndownService from '@joplin/turndown';
-import * as TurndownPluginGfm from '@joplin/turndown-plugin-gfm';
-import { getBasename } from './basename';
-
-// CJS/ESM interop: .default exists in Vite, not in NestJS
-const TurndownService = (_TurndownService as any).default || _TurndownService;
-
-function sanitizeMdLinkText(value: string): string {
-  return value
-    .replace(/\\/g, '\\\\')
-    .replace(/([\[\]!])/g, '\\$1')
-    .replace(/[\r\n]+/g, ' ');
-}
-
-// Tags turndown treats as void (self-closing). Footnote references render as an
-// empty <sup data-footnote-ref> whose meaning lives entirely in its data-id;
-// without marking it void, turndown's blank-node removal drops it before our
-// rule runs, losing the `[^id]` marker. Mirrors turndown's built-in list.
-const TURNDOWN_VOID_ELEMENTS = [
-  'AREA', 'BASE', 'BR', 'COL', 'COMMAND', 'EMBED', 'HR', 'IMG', 'INPUT',
-  'KEYGEN', 'LINK', 'META', 'PARAM', 'SOURCE', 'TRACK', 'WBR',
-];
-
-function isVoidNode(node: any): boolean {
-  const name = node?.nodeName?.toUpperCase?.();
-  if (!name) return false;
-  if (name === 'SUP' && node.hasAttribute?.('data-footnote-ref')) {
-    return true;
-  }
-  return TURNDOWN_VOID_ELEMENTS.indexOf(name) !== -1;
-}
-
-/**
- * An empty <sup data-footnote-ref> is "blank" to turndown, which removes blank
- * inline nodes (RootNode/Node use a module-level isVoid the options cannot
- * override). To survive, inject the id as text content so the node is non-blank;
- * the footnoteReference rule then reads data-id and emits `[^id]`.
- */
-function fillEmptyFootnoteRefs(html: string): string {
-  return html.replace(
-    /<sup\b([^>]*\bdata-footnote-ref\b[^>]*)>\s*<\/sup>/gi,
-    (_m, attrs) => `<sup${attrs}>​</sup>`,
-  );
-}
-
-/**
- * `pageBreak` and `transclusionReference` are childless atom <div>s. Like an
- * empty footnote ref (see above), turndown treats a childless block as "blank"
- * and replaces it with the blankRule BEFORE any custom rule can fire — so the
- * node disappears from the export with no trace (#206 mdrt-2). Inject a
- * zero-width space so the node is non-blank and our lossless rule runs; the
- * rule rebuilds the tag from the element's attributes, so the injected char
- * never reaches the output.
- */
-function fillEmptyAtomBlocks(html: string): string {
-  return html.replace(
-    /<div\b([^>]*\bdata-type="(?:pageBreak|transclusionReference)"[^>]*)>\s*<\/div>/gi,
-    (_m, attrs) => `<div${attrs}>​</div>`,
-  );
-}
-
-/** HTML-escape an attribute value so a re-emitted raw-HTML tag is well-formed. */
-function escapeHtmlAttr(value: string): string {
-  return value
-    .replace(/&/g, '&amp;')
-    .replace(/"/g, '&quot;')
-    .replace(/</g, '&lt;')
-    .replace(/>/g, '&gt;');
-}
-
-/** HTML-escape text placed inside a re-emitted raw-HTML element. */
-function escapeHtmlText(value: string): string {
-  return value
-    .replace(/&/g, '&amp;')
-    .replace(/</g, '&lt;')
-    .replace(/>/g, '&gt;');
-}
-
-/**
- * Serialize ALL of an element's attributes back to a raw-HTML attribute string
- * (leading space included). Generic on purpose: a custom node's identity lives
- * entirely in its `data-*` attributes (data-id, data-color, data-source-page-id,
- * data-transclusion-id, …), and serializing every attribute keeps the export
- * lossless regardless of which attributes a given node carries.
- */
-function serializeAttrs(node: any): string {
-  const attrs = node?.attributes;
-  if (!attrs) return '';
-  return Array.from(attrs as ArrayLike<{ name: string; value: string }>)
-    .map((attr) => ` ${attr.name}="${escapeHtmlAttr(attr.value ?? '')}"`)
-    .join('');
-}
-
-export function htmlToMarkdown(html: string): string {
-  const turndownService = new TurndownService({
-    headingStyle: 'atx',
-    codeBlockStyle: 'fenced',
-    hr: '---',
-    bulletListMarker: '-',
-    isVoid: isVoidNode,
-  });
-
-  turndownService.use([
-    TurndownPluginGfm.tables,
-    TurndownPluginGfm.strikethrough,
-    TurndownPluginGfm.highlightedCodeBlock,
-    taskList,
-    callout,
-    preserveDetail,
-    listParagraph,
-    orderedListItem,
-    mathInline,
-    mathBlock,
-    iframeEmbed,
-    htmlEmbed,
-    spoiler,
-    image,
-    video,
-    footnoteReference,
-    footnotesList,
-    pageBreak,
-    transclusionReference,
-    mention,
-    status,
-  ]);
-  return turndownService
-    .turndown(fillEmptyAtomBlocks(fillEmptyFootnoteRefs(html)))
-    .replaceAll('<br>', ' ');
-}
-
-/**
- * Lossless export rules for custom nodes that have NO native Markdown syntax
- * (#206 mdrt-2). Markdown cannot represent a page break, a transclusion
- * reference, a mention's stable id, or a status chip's color — so rather than
- * letting turndown silently drop them, each rule re-emits the node as raw HTML
- * carrying every `data-*` attribute. Plain-Markdown viewers ignore the inert
- * tag, and the import path round-trips it: `markdownToHtml` passes raw HTML
- * through and each node's `parseHTML` (`div[data-type="…"]`, `span[…]`) rebuilds
- * the ProseMirror node with its attributes intact.
- */
-function pageBreak(turndownService: _TurndownService) {
-  turndownService.addRule('pageBreak', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'DIV' &&
-        node.getAttribute('data-type') === 'pageBreak'
-      );
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      return `\n\n<div${serializeAttrs(node)}></div>\n\n`;
-    },
-  });
-}
-
-function transclusionReference(turndownService: _TurndownService) {
-  turndownService.addRule('transclusionReference', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'DIV' &&
-        node.getAttribute('data-type') === 'transclusionReference'
-      );
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      return `\n\n<div${serializeAttrs(node)}></div>\n\n`;
-    },
-  });
-}
-
-function mention(turndownService: _TurndownService) {
-  turndownService.addRule('mention', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'SPAN' &&
-        node.getAttribute('data-type') === 'mention'
-      );
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const text = escapeHtmlText(node.textContent || '');
-      return `<span${serializeAttrs(node)}>${text}</span>`;
-    },
-  });
-}
-
-function status(turndownService: _TurndownService) {
-  turndownService.addRule('status', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'SPAN' && node.getAttribute('data-type') === 'status'
-      );
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const text = escapeHtmlText(node.textContent || '');
-      return `<span${serializeAttrs(node)}>${text}</span>`;
-    },
-  });
-}
-
-/**
- * Serialize the `htmlEmbed` node to Markdown.
- *
- * Markdown has no native representation for an arbitrary-HTML block, so we
- * preserve the node losslessly as an HTML comment carrying the base64-encoded
- * source (the same `data-source` payload the node stores). `markdownToHtml`
- * recognizes the same marker and rebuilds the node, so the round-trip
- * MD -> HTML -> JSON keeps the source intact. The comment also keeps the raw
- * markup inert in the exported `.md` file (it does not render in plain Markdown
- * viewers).
- */
-function htmlEmbed(turndownService: _TurndownService) {
-  turndownService.addRule('htmlEmbed', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'DIV' &&
-        node.getAttribute('data-type') === 'htmlEmbed'
-      );
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const encoded = node.getAttribute('data-source') || '';
-      return `\n\n<!--html-embed:${encoded}-->\n\n`;
-    },
-  });
-}
-
-/**
- * Serialize the `spoiler` inline mark to lossless raw inline HTML.
- *
- * Markdown has no native spoiler syntax, so we emit the same `<span
- * data-spoiler="true">…</span>` the mark renders. `marked` passes inline raw HTML
- * through untouched, and `generateJSON` restores the mark via its parseHTML, so
- * the round-trip MD -> HTML -> JSON keeps the spoiler intact. The UI-only
- * `is-revealed` state is never serialized.
- */
-function spoiler(turndownService: _TurndownService) {
-  turndownService.addRule('spoiler', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'SPAN' &&
-        node.getAttribute('data-spoiler') === 'true'
-      );
-    },
-    replacement: function (content: string) {
-      return `<span data-spoiler="true">${content}</span>`;
-    },
-  });
-}
-
-function listParagraph(turndownService: _TurndownService) {
-  turndownService.addRule('paragraph', {
-    filter: ['p'],
-    replacement: (content: string, node: HTMLInputElement) => {
-      if (node.parentElement?.nodeName === 'LI') {
-        return content;
-      }
-      return `\n\n${content}\n\n`;
-    },
-  });
-}
-
-function orderedListItem(turndownService: _TurndownService) {
-  turndownService.addRule('orderedListItem', {
-    filter: function (node: HTMLInputElement) {
-      return node.nodeName === 'LI' && node.getAttribute('data-type') !== 'taskItem';
-    },
-    replacement: (content: string, node: HTMLInputElement, options: any) => {
-      const parent = node.parentNode as HTMLElement;
-      if (parent.nodeName !== 'OL' && parent.nodeName !== 'UL') {
-        return content;
-      }
-
-      content = content
-        .replace(/^\n+/, '')
-        .replace(/\n+$/, '\n')
-        .replace(/\n/gm, '\n  ');
-
-      let prefix: string;
-      if (parent.nodeName === 'OL') {
-        const start = parseInt(parent.getAttribute('start') || '1', 10);
-        const index = Array.prototype.indexOf.call(parent.children, node);
-        prefix = `${start + index}. `;
-      } else {
-        prefix = `${options.bulletListMarker} `;
-      }
-
-      return (
-        prefix +
-        content +
-        (node.nextSibling && !/\n$/.test(content) ? '\n' : '')
-      );
-    },
-  });
-}
-
-function callout(turndownService: _TurndownService) {
-  turndownService.addRule('callout', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'DIV' && node.getAttribute('data-type') === 'callout'
-      );
-    },
-    replacement: function (content: string, node: HTMLInputElement) {
-      const calloutType = node.getAttribute('data-callout-type');
-      return `\n\n:::${calloutType}\n${content.trim()}\n:::\n\n`;
-    },
-  });
-}
-
-function taskList(turndownService: _TurndownService) {
-  turndownService.addRule('taskListItem', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.getAttribute('data-type') === 'taskItem' &&
-        node.parentNode.nodeName === 'UL'
-      );
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const isChecked = node.getAttribute('data-checked') === 'true';
-      const div = node.querySelector('div');
-      const text = div ? div.textContent.trim() : node.textContent.trim();
-
-      const prefix = `- ${isChecked ? '[x]' : '[ ]'} `;
-
-      return (
-        prefix +
-        text +
-        (node.nextSibling && !/\n$/.test(text) ? '\n' : '')
-      );
-    },
-  });
-}
-
-function preserveDetail(turndownService: _TurndownService) {
-  turndownService.addRule('preserveDetail', {
-    filter: function (node: HTMLInputElement) {
-      return node.nodeName === 'DETAILS';
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const summary = node.querySelector(':scope > summary');
-      let detailSummary = '';
-
-      if (summary) {
-        detailSummary = `<summary>${turndownService.turndown(summary.innerHTML)}</summary>`;
-      }
-
-      const detailsContent = Array.from(node.childNodes)
-        .filter((child) => child.nodeName !== 'SUMMARY')
-        .map((child) =>
-          child.nodeType === 1
-            ? turndownService.turndown((child as HTMLElement).outerHTML)
-            : child.textContent,
-        )
-        .join('');
-
-      return `\n<details>\n${detailSummary}\n\n${detailsContent}\n\n</details>\n`;
-    },
-  });
-}
-
-function mathInline(turndownService: _TurndownService) {
-  turndownService.addRule('mathInline', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'SPAN' &&
-        node.getAttribute('data-type') === 'mathInline'
-      );
-    },
-    replacement: function (content: string) {
-      return `$${content}$`;
-    },
-  });
-}
-
-function mathBlock(turndownService: _TurndownService) {
-  turndownService.addRule('mathBlock', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'DIV' &&
-        node.getAttribute('data-type') === 'mathBlock'
-      );
-    },
-    replacement: function (content: string) {
-      return `\n$$\n${content}\n$$\n`;
-    },
-  });
-}
-
-function iframeEmbed(turndownService: _TurndownService) {
-  turndownService.addRule('iframeEmbed', {
-    filter: function (node: HTMLInputElement) {
-      return node.nodeName === 'IFRAME';
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const src = node.getAttribute('src');
-      return '[' + src + '](' + src + ')';
-    },
-  });
-}
-
-function image(turndownService: _TurndownService) {
-  turndownService.addRule('image', {
-    filter: 'img',
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const src = node.getAttribute('src') || '';
-      if (!src) return '';
-      const caption = node.getAttribute('data-caption') || '';
-      if (caption) {
-        // ![]() can't carry a caption, so emit a raw <img> wrapped in a block
-        // <div>. marked passes it through and the image extension's parseHTML
-        // restores the caption from data-caption.
-        const parts = [`src="${escapeHtmlAttr(src)}"`];
-        const alt = node.getAttribute('alt') || '';
-        if (alt) parts.push(`alt="${escapeHtmlAttr(alt)}"`);
-        parts.push(`data-caption="${escapeHtmlAttr(caption)}"`);
-        return `<div><img ${parts.join(' ')}></div>`;
-      }
-      const alt = sanitizeMdLinkText(node.getAttribute('alt') || '');
-      const title = node.getAttribute('title') || '';
-      const titlePart = title ? ' "' + title.replace(/"/g, '\\"') + '"' : '';
-      return '![' + alt + '](' + src + titlePart + ')';
-    },
-  });
-}
-
-/**
- * Footnote reference (inline atom) -> pandoc/GFM marker `[^id]`.
- * The visible number is derived (not stored), so the id is the stable anchor.
- */
-function footnoteReference(turndownService: _TurndownService) {
-  turndownService.addRule('footnoteReference', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'SUP' && node.hasAttribute('data-footnote-ref')
-      );
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const id = node.getAttribute('data-id') || '';
-      return id ? `[^${id}]` : '';
-    },
-  });
-}
-
-/**
- * Footnotes container -> the list of `[^id]: text` definitions at the end of
- * the document (one per line). Each footnoteDefinition inside emits its own
- * `[^id]: ...` line; turndown joins them with the surrounding block spacing.
- */
-function footnotesList(turndownService: _TurndownService) {
-  turndownService.addRule('footnoteDefinition', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'DIV' && node.hasAttribute('data-footnote-def')
-      );
-    },
-    replacement: function (content: string, node: HTMLInputElement) {
-      const id = node.getAttribute('data-id') || '';
-      // Collapse internal newlines so the definition stays a single MD line;
-      // continuation lines are a v2 refinement.
-      const text = content.replace(/\s*\n+\s*/g, ' ').trim();
-      return id ? `\n[^${id}]: ${text}\n` : '';
-    },
-  });
-
-  turndownService.addRule('footnotesList', {
-    filter: function (node: HTMLInputElement) {
-      return (
-        node.nodeName === 'SECTION' && node.hasAttribute('data-footnotes')
-      );
-    },
-    replacement: function (content: string) {
-      return `\n\n${content.trim()}\n`;
-    },
-  });
-}
-
-function video(turndownService: _TurndownService) {
-  turndownService.addRule('video', {
-    filter: function (node: HTMLInputElement) {
-      return node.tagName === 'VIDEO';
-    },
-    replacement: function (_content: string, node: HTMLInputElement) {
-      const src = node.getAttribute('src') || '';
-      const ariaLabel = node.getAttribute('aria-label');
-      const name = sanitizeMdLinkText(
-        ariaLabel || getBasename(src) || src,
-      );
-      return '[' + name + '](' + src + ')';
-    },
-  });
-}

packages/editor-ext/vitest.config.ts

@@ -14,10 +14,15 @@ export default defineConfig({
       provider: "v8",
       reporter: ["text-summary", "text"],
       all: false,
+      // functions lowered 60 -> 57 after issue #347 removed the editor-ext
+      // markdown layer (src/lib/markdown) and its image/footnote round-trip
+      // specs: that markdown behavior now lives in — and is tested by —
+      // @docmost/prosemirror-markdown, so the editor-ext baseline shifts down.
+      // Still a real gate (a few points below the post-removal measured level).
       thresholds: {
         statements: 54,
         branches: 44,
-        functions: 60,
+        functions: 57,
         lines: 54,
       },
     },

packages/mcp/README.md

@@ -40,7 +40,7 @@ There are several Docmost MCPs. Here is a capability-by-capability comparison.
 | **Enterprise license required** | **No** | **Yes** | No | No | No |
 | Authentication | email + password, **auto re-auth** | API key | email + password | cookie `authToken` (copy from DevTools) | Docmost API / **direct PostgreSQL** |
 | Read page as Markdown | ✅ | ✅ | ✅ | ✅ | ✅ (read-only) |
-| **Lossless Markdown round-trip** (export / import, keeps comment anchors) | ✅ | — | — | — | — |
+| **Markdown round-trip** (export / import, keeps comment anchors) | ✅ | — | — | — | — |
 | Read **lossless ProseMirror JSON** (with block ids) | ✅ | — | — | — | — |
 | **Compact page outline** (cheap block-id lookup) | ✅ | — | — | — | — |
 | **Fetch a single block** (by id or index) | ✅ | — | — | — | — |
@@ -115,8 +115,10 @@ All 41 tools, grouped by what you'd reach for them.
 - **`listPages`** — Recent pages in a space, ordered by `updatedAt` desc (default 50,
   max 100). Use `search` for lookups in large spaces.
 - **`search`** — Full-text search across pages and content (bounded by `limit`, max 100).
-- **`getPage`** — A page's content as clean **Markdown** (convenient, but a *lossy*
+- **`getPage`** — A page's content as clean **Markdown** (canonical for text; drops only
-  view — block ids and exact table/callout structure are approximated).
+  block ids, resolved-comment anchors, and a fixed no-Markdown-representation attr set —
+  table spans/colwidth/background, indent, `callout.icon`, `orderedList.type`, and link
+  `internal`/`target`/`rel`/`class`; use `getPageJson` when you need those).
 - **`getPageJson`** — A page's **lossless ProseMirror/TipTap JSON**, including every
   block's `attrs.id` and the `slugId` used in URLs. This is what the per-block editing
   tools consume.
@@ -186,10 +188,14 @@ All 41 tools, grouped by what you'd reach for them.
 
 ### Markdown round-trip
 
-- **`exportPageMarkdown`** — Export a page to a single self-contained, **lossless
+- **`exportPageMarkdown`** — Export a page to a single self-contained
-  Docmost-flavoured Markdown** file: a meta header, the body with inline comment anchors
+  **Docmost-flavoured Markdown** file: a meta header, the body with inline comment anchors
-  and diagrams, and a trailing comments-thread block. To replace a page's body from plain
+  and diagrams, and a trailing comments-thread block. The download → edit → import
-  authoring Markdown, use `updatePageMarkdown`.
+  round-trip regenerates block ids and **silently drops** the no-Markdown-representation
+  attr set (table merge spans/colwidth/background, indent, `callout.icon`,
+  `orderedList.type`, link `internal`/`target`/`rel`/`class`); keep those in ProseMirror
+  JSON if they must survive. To replace a page's body from plain authoring Markdown, use
+  `updatePageMarkdown`.
 
 > **Removed in this release:** `importPageMarkdown` (the round-trip parser for an
 > exported Docmost-Markdown file) is **no longer exposed on the external MCP surface**.
@@ -287,21 +293,42 @@ so capable clients steer the model automatically.
   the debounced REST snapshot), then **reads → transforms → writes synchronously** in one
   tick so no remote update can interleave, and **waits for persistence acknowledgement**
   before returning.
-- **Per-page write serialization.** A per-`pageId` async mutex ensures two MCP writes to
+- **Per-page write serialization.** A per-`pageId` async mutex (keyed by the resolved
-  the same page never overlap; different pages never block each other.
+  page **UUID**, never a slugId) ensures two MCP writes to the same page never overlap;
+  different pages never block each other. The lock helper fails fast if it is ever handed
+  a non-UUID key, so a write path that forgot to resolve the id can never silently lock
+  under a split key.
+
+  **Deploy requirement — single instance or sticky sessions.** This mutex is an
+  in-process `Map`, and the cached collab sessions and the `stash_page` blob store are
+  RAM-only and process-local. Behind a **multi-replica** load balancer **without sticky
+  sessions**, two replicas can each "hold" the lock for the same page at once and per-page
+  serialization is silently lost. Run the MCP/app as a **single instance**, or pin each
+  page's traffic to one replica (sticky sessions / consistent hashing on the page id).
+  There is deliberately no cross-process (e.g. Postgres advisory) lock yet — a conscious
+  documented constraint. See the `Dockerfile` comment and the `MCP collaboration write
+  path` block in `.env.example`.
+
+  **Rights-staleness window.** A cached collab session writes under the token captured at
+  connect time (and the collab-token cache reuses a token for its TTL), so a **revoked**
+  page access can lag by up to `MCP_COLLAB_SESSION_MAX_AGE_MS` (the hard session lifetime,
+  default 10 min) before the next re-auth picks it up. Lower it to shorten the lag at the
+  cost of more reconnects. This bounded window is an accepted trade-off; there is no
+  push-based cache invalidation on a rights change.
 - **Transparent re-authentication.** Login uses email/password; expired tokens are
   refreshed automatically on the first 401/403 (covering JSON, multipart upload, and the
   collaboration-token path), with in-flight login de-duplication so a burst of calls
   triggers a single re-login.
-- **Lossless and lossy reads.** `getPageJson` returns the exact ProseMirror tree with
+- **Precise reads.** `getPageJson` returns the exact ProseMirror tree with block ids;
-  block ids; `getPage` returns clean Markdown for convenience.
+  `getPage` returns canonical Markdown that drops only a fixed, documented attr set.
 - **Full Docmost schema.** Markdown↔ProseMirror conversion supports callouts (including
   nested), task lists (bullet *and* numbered checklists), tables, math blocks, embeds,
   highlights, sub/superscript and more, with defensive caps against pathological input.
-- **Structured tables & lossless Markdown round-trip.** Tables can be edited as a matrix
+- **Structured tables & Markdown round-trip.** Tables can be edited as a matrix
   (read, insert/delete rows, set cells by `[row,col]`) without resending the document, and
   a page can be exported to and re-imported from a self-contained Docmost-flavoured
-  Markdown file that preserves inline comment anchors and diagrams.
+  Markdown file that preserves inline comment anchors and diagrams (block ids regenerate
+  and a fixed no-Markdown-representation attr set is dropped — see `exportPageMarkdown`).
 - **Token-optimized responses.** API responses are filtered down to the fields agents
   actually need, and large collections (spaces, pages, comments, history) are paginated.
 - **Hardened runtime.** Global handlers keep a stray socket error from tearing down the

packages/mcp/README.ru.md

@@ -43,7 +43,7 @@ Docmost-MCP не сочетают:
 | **Нужна enterprise-лицензия** | **Нет** | **Да** | Нет | Нет | Нет |
 | Аутентификация | email + пароль, **авто-переавторизация** | API-ключ | email + пароль | cookie `authToken` (копировать из DevTools) | API Docmost / **напрямую PostgreSQL** |
 | Чтение страницы как Markdown | ✅ | ✅ | ✅ | ✅ | ✅ (только чтение) |
-| **Lossless Markdown round-trip** (экспорт/импорт, сохраняет якоря комментариев) | ✅ | — | — | — | — |
+| **Markdown round-trip** (экспорт/импорт, сохраняет якоря комментариев) | ✅ | — | — | — | — |
 | Чтение **lossless ProseMirror JSON** (с id блоков) | ✅ | — | — | — | — |
 | **Компактная структура страницы** (дешёвый поиск id блока) | ✅ | — | — | — | — |
 | **Получение одного блока** (по id или индексу) | ✅ | — | — | — | — |
@@ -119,8 +119,11 @@ Docmost-MCP не сочетают:
   50, максимум 100). Для поиска в больших пространствах используйте `search`.
 - **`search`** — Полнотекстовый поиск по страницам и контенту (ограничен `limit`, максимум
   100).
-- **`getPage`** — Контент страницы как чистый **Markdown** (удобно, но это
+- **`getPage`** — Контент страницы как чистый **Markdown** (канонично для текста; теряет
-  *lossy*-представление — id блоков и точная структура таблиц/коллаутов аппроксимируются).
+  лишь id блоков, якоря разрешённых комментариев и фиксированный набор атрибутов без
+  markdown-представления — спаны/colwidth/фон ячеек таблиц, отступы (indent),
+  `callout.icon`, `orderedList.type` и `internal`/`target`/`rel`/`class` у ссылок;
+  используйте `getPageJson`, когда они нужны).
 - **`getPageJson`** — **Lossless ProseMirror/TipTap JSON** страницы, включая `attrs.id`
   каждого блока и `slugId`, используемый в URL. Именно его потребляют инструменты
   поблочного редактирования.
@@ -191,10 +194,14 @@ Docmost-MCP не сочетают:
 
 ### Markdown: экспорт и импорт
 
-- **`exportPageMarkdown`** — Экспортировать страницу в один самодостаточный, **lossless
+- **`exportPageMarkdown`** — Экспортировать страницу в один самодостаточный
-  Markdown в диалекте Docmost**: мета-заголовок, тело с inline-якорями комментариев и
+  **Markdown в диалекте Docmost**: мета-заголовок, тело с inline-якорями комментариев и
-  диаграммами и завершающий блок тредов комментариев. Чтобы заменить тело страницы из
+  диаграммами и завершающий блок тредов комментариев. Round-trip скачать → отредактировать →
-  обычного авторского Markdown, используйте `updatePageMarkdown`.
+  импортировать перегенерирует id блоков и **молча отбрасывает** набор атрибутов без
+  markdown-представления (спаны/colwidth/фон ячеек таблиц, отступы (indent), `callout.icon`,
+  `orderedList.type`, `internal`/`target`/`rel`/`class` у ссылок); держите их в ProseMirror
+  JSON, если они должны выжить. Чтобы заменить тело страницы из обычного авторского Markdown,
+  используйте `updatePageMarkdown`.
 
 > **Удалено в этом релизе:** `importPageMarkdown` (парсер round-trip для
 > экспортированного Docmost-Markdown-файла) **больше не отдаётся на внешней MCP-поверхности**.
@@ -295,23 +302,44 @@ Docmost-MCP не сочетают:
   правки, которых ещё нет в дебаунс-снапшоте REST), затем **читает → трансформирует →
   пишет синхронно** в одном тике, чтобы никакое удалённое обновление не вклинилось, и
   **ждёт подтверждения сохранения** до возврата.
-- **Сериализация записи по странице.** Асинхронный мьютекс по `pageId` гарантирует, что
+- **Сериализация записи по странице.** Асинхронный мьютекс по разрешённому **UUID**
-  две записи MCP в одну страницу никогда не пересекаются; разные страницы друг друга не
+  страницы (никогда не по slugId) гарантирует, что две записи MCP в одну страницу никогда
-  блокируют.
+  не пересекаются; разные страницы друг друга не блокируют. Хелпер блокировки падает сразу
+  (fail-fast), если ему передали не-UUID ключ, — путь записи, забывший разрезолвить id, не
+  сможет молча взять лок под расщеплённым ключом.
+
+  **Требование к деплою — один инстанс или sticky-сессии.** Этот мьютекс — процесс-локальный
+  `Map`, а кэш collab-сессий и хранилище `stashPage` живут только в RAM одного процесса. За
+  **мультиреплика**-балансировщиком **без sticky-сессий** две реплики могут одновременно
+  «держать» лок одной страницы, и сериализация по странице молча теряется. Запускайте
+  MCP/приложение **одним инстансом** либо прибивайте трафик страницы к одной реплике
+  (sticky-сессии / consistent hashing по id страницы). Кросс-процессной блокировки (например,
+  Postgres advisory-lock) намеренно пока нет — осознанное задокументированное ограничение.
+  См. комментарий в `Dockerfile` и блок `MCP collaboration write path` в `.env.example`.
+
+  **Окно устаревших прав.** Кэшированная collab-сессия пишет под токеном, захваченным в
+  момент connect (а кэш collab-токена переиспользует токен в пределах своего TTL), поэтому
+  **отозванный** доступ к странице может лагать до `MCP_COLLAB_SESSION_MAX_AGE_MS` (жёсткий
+  срок жизни сессии, по умолчанию 10 мин), пока следующая переавторизация его не подхватит.
+  Уменьшите значение, чтобы сократить лаг ценой большего числа переподключений. Это
+  ограниченное окно — принятый trade-off; push-инвалидации кэша при смене прав нет.
 - **Прозрачная переавторизация.** Логин по email/паролю; истёкшие токены обновляются
   автоматически на первом 401/403 (покрывая JSON, multipart-загрузку и путь токена
   коллаборации), с дедупликацией параллельных логинов, так что пачка вызовов вызывает один
   повторный логин.
-- **Lossless- и lossy-чтение.** `getPageJson` возвращает точное дерево ProseMirror с id
+- **Точные чтения.** `getPageJson` возвращает точное дерево ProseMirror с id блоков;
-  блоков; `getPage` возвращает чистый Markdown для удобства.
+  `getPage` возвращает канонический Markdown, теряющий лишь фиксированный, документированный
+  набор атрибутов.
 - **Полная схема Docmost.** Конвертация Markdown↔ProseMirror поддерживает коллауты
   (включая вложенные), списки задач (маркированные *и* нумерованные чек-листы), таблицы,
   блоки формул, эмбеды, выделение, под/надстрочный текст и прочее, с защитными лимитами
   против патологического ввода.
-- **Структурные таблицы и lossless Markdown round-trip.** Таблицы можно редактировать как
+- **Структурные таблицы и Markdown round-trip.** Таблицы можно редактировать как
   матрицу (чтение, вставка/удаление строк, задание ячеек по `[row, col]`) без пересылки
   документа, а страницу — экспортировать и заново импортировать как самодостаточный
-  Markdown-файл в диалекте Docmost, сохраняющий inline-якоря комментариев и диаграммы.
+  Markdown-файл в диалекте Docmost, сохраняющий inline-якоря комментариев и диаграммы
+  (id блоков перегенерируются, а фиксированный набор атрибутов без markdown-представления
+  отбрасывается — см. `exportPageMarkdown`).
 - **Ответы, оптимизированные по токенам.** Ответы API урезаются до полей, действительно
   нужных агентам, а большие коллекции (пространства, страницы, комментарии, история)
   пагинируются.

packages/mcp/data/drawio-presets.json

@@ -0,0 +1,63 @@
+{
+  "$comment": "Semantic palettes for drawioFromGraph (issue #425). DATA, not code: node `kind` -> fill/stroke slot, edge `kind` -> line-style props, per preset. The `default` node palette is the issue's base table; `dark` keeps the same hues on a dark canvas with lighter strokes/font; `colorblind-safe` maps every slot onto the Okabe-Ito qualitative palette (8 colours proven distinguishable for all common colour-vision deficiencies) so no two adjacent kinds collide. `fontColor`/`fillColor`/`strokeColor` are exact draw.io values. `edgeDefault` is the fallback line style; `group` is the (always-transparent) container stroke per preset.",
+  "presets": {
+    "default": {
+      "canvasDark": false,
+      "nodes": {
+        "service":  { "fillColor": "#dae8fc", "strokeColor": "#6c8ebf", "fontColor": "#000000" },
+        "db":       { "fillColor": "#d5e8d4", "strokeColor": "#82b366", "fontColor": "#000000" },
+        "queue":    { "fillColor": "#fff2cc", "strokeColor": "#d6b656", "fontColor": "#000000" },
+        "gateway":  { "fillColor": "#ffe6cc", "strokeColor": "#d79b00", "fontColor": "#000000" },
+        "error":    { "fillColor": "#f8cecc", "strokeColor": "#b85450", "fontColor": "#000000" },
+        "external": { "fillColor": "#f5f5f5", "strokeColor": "#666666", "fontColor": "#333333" },
+        "security": { "fillColor": "#e1d5e7", "strokeColor": "#9673a6", "fontColor": "#000000" }
+      },
+      "edges": {
+        "sync":  { "props": "" },
+        "async": { "props": "dashed=1;" },
+        "error": { "props": "dashed=1;strokeColor=#DD344C;" }
+      },
+      "edgeDefault": { "strokeColor": "#333333", "fontColor": "#333333" },
+      "group": { "strokeColor": "#666666", "fontColor": "#333333" }
+    },
+    "dark": {
+      "canvasDark": true,
+      "nodes": {
+        "service":  { "fillColor": "#1a2a44", "strokeColor": "#7ea6e0", "fontColor": "#dae8fc" },
+        "db":       { "fillColor": "#1f331e", "strokeColor": "#97d077", "fontColor": "#d5e8d4" },
+        "queue":    { "fillColor": "#3a3218", "strokeColor": "#e5c15a", "fontColor": "#fff2cc" },
+        "gateway":  { "fillColor": "#3a2812", "strokeColor": "#ffb570", "fontColor": "#ffe6cc" },
+        "error":    { "fillColor": "#3a1c1b", "strokeColor": "#e08e8b", "fontColor": "#f8cecc" },
+        "external": { "fillColor": "#2b2b2b", "strokeColor": "#999999", "fontColor": "#e0e0e0" },
+        "security": { "fillColor": "#2c2338", "strokeColor": "#b39ddb", "fontColor": "#e1d5e7" }
+      },
+      "edges": {
+        "sync":  { "props": "" },
+        "async": { "props": "dashed=1;" },
+        "error": { "props": "dashed=1;strokeColor=#ff6b6b;" }
+      },
+      "edgeDefault": { "strokeColor": "#cccccc", "fontColor": "#e0e0e0" },
+      "group": { "strokeColor": "#aaaaaa", "fontColor": "#e0e0e0" }
+    },
+    "colorblind-safe": {
+      "canvasDark": false,
+      "okabeIto": ["#000000", "#E69F00", "#56B4E9", "#009E73", "#F0E442", "#0072B2", "#D55E00", "#CC79A7"],
+      "nodes": {
+        "service":  { "fillColor": "#D6E9F5", "strokeColor": "#0072B2", "fontColor": "#000000" },
+        "db":       { "fillColor": "#D6EFE4", "strokeColor": "#009E73", "fontColor": "#000000" },
+        "queue":    { "fillColor": "#FCF8CC", "strokeColor": "#F0E442", "fontColor": "#000000" },
+        "gateway":  { "fillColor": "#FBEBD0", "strokeColor": "#E69F00", "fontColor": "#000000" },
+        "error":    { "fillColor": "#F7DDCC", "strokeColor": "#D55E00", "fontColor": "#000000" },
+        "external": { "fillColor": "#EDEDED", "strokeColor": "#000000", "fontColor": "#000000" },
+        "security": { "fillColor": "#F3DEEB", "strokeColor": "#CC79A7", "fontColor": "#000000" }
+      },
+      "edges": {
+        "sync":  { "props": "" },
+        "async": { "props": "dashed=1;" },
+        "error": { "props": "dashed=1;strokeColor=#D55E00;" }
+      },
+      "edgeDefault": { "strokeColor": "#000000", "fontColor": "#000000" },
+      "group": { "strokeColor": "#000000", "fontColor": "#000000" }
+    }
+  }
+}

packages/mcp/src/client/comments.ts

@@ -0,0 +1,704 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import { assertFullUuid } from "./errors.js";
+import {
+  filterWorkspace,
+  filterSpace,
+  filterPage,
+  filterComment,
+  filterSearchResult,
+} from "../lib/filters.js";
+import { convertProseMirrorToMarkdown } from "../lib/markdown-converter.js";
+import {
+  updatePageContentRealtime,
+  replacePageContent,
+  markdownToProseMirror,
+  markdownToProseMirrorCanonical,
+  mutatePageContent,
+  assertYjsEncodable,
+  MutationResult,
+} from "../lib/collaboration.js";
+import {
+  replaceNodeById,
+  replaceNodeByIdWithMany,
+  reassignCollidingBlockIds,
+  deleteNodeById,
+  assertUnambiguousMatch,
+  insertNodeRelative,
+  insertNodesRelative,
+  blockPlainText,
+  buildOutline,
+  getNodeByRef,
+  readTable,
+  insertTableRow,
+  deleteTableRow,
+  updateTableCell,
+  findInvalidNode,
+} from "@docmost/prosemirror-markdown";
+import {
+  applyAnchorInDoc,
+  countAnchorMatches,
+  getAnchoredText,
+  resolveAnchorSelection,
+  normalizeForMatch,
+} from "../lib/comment-anchor.js";
+import { closestBlockHint } from "../lib/text-normalize.js";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+
+// Public method surface of CommentsMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements ICommentsMixin` fails to compile on drift.
+export interface ICommentsMixin {
+  listComments(pageId: string, includeResolved?: boolean): any;
+  getComment(commentId: string): any;
+  createComment(pageId: string, content: string, type?: "page" | "inline", selection?: string, parentCommentId?: string, suggestedText?: string): any;
+  updateComment(commentId: string, content: string): any;
+  deleteComment(commentId: string): any;
+  resolveComment(commentId: string, resolved: boolean): any;
+  checkNewComments(spaceId: string, since: string, parentPageId?: string): any;
+}
+
+export function CommentsMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & ICommentsMixin> & TBase {
+  abstract class CommentsMixin extends Base implements ICommentsMixin {
+  // --- Comment methods (ported from upstream PR #3 by Max Nikitin) ---
+
+  /**
+   * Normalize a comment's `content` into a ProseMirror doc object before
+   * markdown conversion. createComment/updateComment send content as a
+   * JSON.stringify(...) STRING, and the server stores it as-is, so on read it
+   * comes back as a string. convertProseMirrorToMarkdown returns "" for a
+   * string, so parse it first (guarded — fall back to the raw value on any
+   * parse failure so a non-JSON legacy value is still handled gracefully).
+   */
+  protected parseCommentContent(content: any): any {
+    if (typeof content !== "string") return content;
+    try {
+      return JSON.parse(content);
+    } catch {
+      return content;
+    }
+  }
+
+  /**
+   * List comments on a page (cursor-paginated), content as markdown.
+   *
+   * DEFAULT (`includeResolved = false`) hides RESOLVED THREADS WHOLESALE so the
+   * agent sees only active discussions: a top-level comment with `resolvedAt`
+   * set AND every reply under it (a reply of a closed thread is part of the
+   * closed thread) are dropped from `items`. `resolvedThreadsHidden` reports how
+   * many resolved top-level threads were hidden so the agent can re-query with
+   * `includeResolved: true` to see everything. Active threads always stay.
+   *
+   * Returns `{ items, resolvedThreadsHidden }` (NOT a bare array) — callers that
+   * need the full feed (lossless export, transformPage, checkNewComments) pass
+   * `includeResolved: true` and read `.items`.
+   */
+  async listComments(pageId: string, includeResolved = false) {
+    await this.ensureAuthenticated();
+    let allComments: any[] = [];
+    let cursor: string | null = null;
+
+    // Hard ceiling + immovable-cursor guard (mirrors paginateAll): if /comments
+    // ever stops advancing the cursor (the exact #442 drift scenario) this loop
+    // would otherwise spin forever accumulating duplicates.
+    const MAX_PAGES = 50;
+    let truncated = false;
+
+    for (let page = 0; page < MAX_PAGES; page++) {
+      const payload: Record<string, any> = { pageId, limit: 100 };
+      if (cursor) payload.cursor = cursor;
+
+      const response = await this.client.post("/comments", payload);
+      const data = response.data.data || response.data;
+      const items = data.items || [];
+      allComments = allComments.concat(items);
+
+      // Advance strictly via the server-issued cursor. A missing nextCursor or a
+      // cursor identical to the one we just sent means the end (or a server that
+      // ignores our pagination param) — stop instead of re-fetching page one.
+      const next: string | null = data.meta?.nextCursor || null;
+      if (!next || next === cursor) break;
+      cursor = next;
+
+      // Reaching the ceiling with a still-advancing cursor means truncation.
+      if (page === MAX_PAGES - 1) truncated = true;
+    }
+
+    if (truncated) {
+      console.warn(
+        `listComments: comments for "${pageId}" truncated at the ${MAX_PAGES}-page cap; more pages exist on the server`,
+      );
+    }
+
+    const mapped = allComments.map((comment: any) => {
+      const markdown = comment.content
+        ? convertProseMirrorToMarkdown(
+            this.parseCommentContent(comment.content),
+          )
+        : "";
+      return filterComment(comment, markdown);
+    });
+
+    if (includeResolved) {
+      return { items: mapped, resolvedThreadsHidden: 0 };
+    }
+
+    // Ids of RESOLVED top-level threads (a top-level comment has no
+    // parentCommentId). A whole thread is hidden when its root is resolved.
+    const resolvedRootIds = new Set(
+      mapped
+        .filter((c) => !c.parentCommentId && c.resolvedAt != null)
+        .map((c) => c.id),
+    );
+
+    const items = mapped.filter((c) => {
+      // Hide the resolved root itself and every reply anchored to it. A reply's
+      // own resolvedAt is irrelevant — its membership follows the parent thread.
+      // ASSUMPTION: Docmost's comment model is FLAT — a reply's parentCommentId
+      // always points at the thread ROOT (no reply-of-reply nesting), so a single
+      // level of parent lookup covers a whole thread. If nested replies are ever
+      // introduced, a deep reply of a resolved thread would need a root-walk here.
+      if (!c.parentCommentId) return !resolvedRootIds.has(c.id);
+      return !resolvedRootIds.has(c.parentCommentId);
+    });
+
+    return { items, resolvedThreadsHidden: resolvedRootIds.size };
+  }
+
+
+  async getComment(commentId: string) {
+    // Fail fast (#436): reject a truncated id before any network call.
+    assertFullUuid("get_comment", "commentId", commentId);
+    await this.ensureAuthenticated();
+    const response = await this.client.post("/comments/info", { commentId });
+    const comment = response.data.data || response.data;
+    const markdown = comment.content
+      ? convertProseMirrorToMarkdown(this.parseCommentContent(comment.content))
+      : "";
+    return {
+      data: filterComment(comment, markdown),
+      success: true,
+    };
+  }
+
+  /** Plain text of each TOP-LEVEL block of `doc`, for anchor-failure hints. */
+  protected topLevelBlockTexts(doc: any): string[] {
+    const content = doc && Array.isArray(doc.content) ? doc.content : [];
+    return content
+      .map((b: any) => blockPlainText(b))
+      .filter((t: string) => t.length > 0);
+  }
+
+  /**
+   * True when per-block anchoring failed but the (normalized) selection DOES
+   * appear in the blocks' joined plain text — i.e. it straddles a block
+   * boundary. Blocks are joined with a newline (collapsed to one space by
+   * normalizeForMatch) so a selection whose parts are separated by a paragraph
+   * break still matches. Callers only reach here after single-block anchoring
+   * (incl. the markdown-strip fallback) has already failed.
+   */
+  protected selectionSpansMultipleBlocks(
+    blockTexts: string[],
+    selection: string,
+  ): boolean {
+    const normSel = normalizeForMatch(selection).norm.trim();
+    if (normSel.length === 0) return false;
+    const joined = normalizeForMatch(blockTexts.join("\n")).norm;
+    return joined.indexOf(normSel) !== -1;
+  }
+
+  /**
+   * Build the actionable error for a createComment anchor MISS, porting
+   * editPageText's self-correction affordances: an explicit "spans multiple
+   * blocks" message when the selection straddles a block boundary, otherwise a
+   * "closest block text" hint quoting the block that holds the selection's
+   * longest token. `live` switches the wording between the pre-check (reading the
+   * persisted page) and the post-create live-anchor failure (which rolls back).
+   */
+  protected anchorNotFoundError(
+    doc: any,
+    selection: string,
+    live: boolean,
+  ): Error {
+    const blockTexts = this.topLevelBlockTexts(doc);
+    const rolled = live ? " The comment was rolled back." : "";
+    if (this.selectionSpansMultipleBlocks(blockTexts, selection)) {
+      return new Error(
+        "createComment: the selection spans multiple blocks; anchor on a " +
+          "contiguous fragment within a SINGLE paragraph/block (<=250 chars)." +
+          rolled,
+      );
+    }
+    const where = live ? "in the live document" : "in the page";
+    return new Error(
+      `createComment: could not find the selection text ${where} to anchor ` +
+        "the comment. Provide the EXACT contiguous text from a single " +
+        "paragraph/block (<=250 chars)." +
+        closestBlockHint(blockTexts, selection) +
+        rolled,
+    );
+  }
+
+  /**
+   * Create an inline comment anchored to its `selection` text, or a reply.
+   *
+   * Top-level comments (no `parentCommentId`) are ALWAYS inline and MUST carry a
+   * `selection`: the `type` argument is kept for interface compatibility but the
+   * effective type is coerced to "inline". The selection has to anchor in the
+   * document; if it cannot, the comment is rolled back and an error is thrown so
+   * the caller is forced to supply a proper inline selection rather than leaving
+   * an orphan, unanchored comment behind. Replies (parentCommentId set) inherit
+   * their parent's anchor: they take NO selection and are not anchored.
+   */
+  async createComment(
+    pageId: string,
+    content: string,
+    type: "page" | "inline" = "page",
+    selection?: string,
+    parentCommentId?: string,
+    suggestedText?: string,
+  ) {
+    // Fail fast (#436): a provided parent id must be a full UUID before any
+    // network call. Validate only when truthy — a falsy parentCommentId means
+    // "top-level comment" (mirrors the isReply computation below), not a reply.
+    if (parentCommentId) {
+      assertFullUuid("createComment", "parentCommentId", parentCommentId);
+    }
+    await this.ensureAuthenticated();
+
+    const isReply = !!parentCommentId;
+    const hasSuggestion =
+      suggestedText !== undefined && suggestedText !== null;
+    // Defense in depth mirroring the server DTO/service: a suggested edit rewrites
+    // the exact anchored text, so it is only meaningful on a top-level inline
+    // comment that carries a selection.
+    if (hasSuggestion) {
+      if (isReply) {
+        throw new Error(
+          "createComment: a suggested edit (suggestedText) cannot be attached to a reply; it applies only to a top-level inline comment.",
+        );
+      }
+      if (!selection || !selection.trim()) {
+        throw new Error(
+          "createComment: a suggested edit (suggestedText) requires a 'selection' to anchor and rewrite.",
+        );
+      }
+    }
+    // Only top-level comments are inline-anchored, so they are stored as
+    // "inline". Replies carry no inline selection, so they keep the historical
+    // general ("page") type — both backward-compatible and semantically correct.
+    // The `type` argument is kept for interface compatibility; createComment
+    // normalizes the effective type internally, so callers may pass "inline".
+    const effectiveType: "page" | "inline" = isReply ? "page" : "inline";
+    if (!isReply && (!selection || !selection.trim())) {
+      throw new Error(
+        "createComment: an inline 'selection' (exact text to anchor on) is required for a top-level comment",
+      );
+    }
+
+    // For a SUGGESTION, the value we store as the comment's `selection` must be
+    // the RAW document substring the mark lands on (typographic quotes/dashes,
+    // nbsp, collapsed whitespace), NOT the agent's ASCII input. The anchor is
+    // placed via normalization, so when the doc was auto-converted to
+    // typographic the raw substring differs from the agent input; apply-time
+    // compares the stored selection to the marked doc text STRICTLY, so storing
+    // the raw substring is what makes "Apply" succeed instead of a spurious 409.
+    // Captured in the pre-check below (which already reads the page) and used as
+    // payload.selection. Ordinary comments keep sending the raw agent selection.
+    let anchoredSelection: string | null = null;
+    // Set when the anchor matched only after stripping markdown from the
+    // selection (the strip fallback); surfaced as a soft warning like
+    // editPageText does, so a stale-markdown selection is flagged.
+    let anchorNormalized = false;
+
+    // For a top-level comment, fail BEFORE creating anything when the selection
+    // is not present in the persisted document — this avoids leaving an orphan
+    // comment + notification behind. A read failure (network) is non-fatal: the
+    // live anchor step below still enforces the anchoring invariant.
+    if (!isReply && selection) {
+      try {
+        const page = await this.getPageJson(pageId);
+        if (hasSuggestion) {
+          // A suggestion's anchor MUST be unambiguous: applying it rewrites the
+          // exact anchored text, and ordinary anchoring silently takes the first
+          // occurrence, so 0 matches -> not found and >=2 -> ambiguous, both
+          // rejected BEFORE creating the comment.
+          const matches = countAnchorMatches(page.content, selection);
+          if (matches === 0) {
+            throw this.anchorNotFoundError(page.content, selection, false);
+          }
+          if (matches >= 2) {
+            throw new Error(
+              `createComment: the suggestion's selection is ambiguous — it occurs ${matches} times in the page. ` +
+                "A suggested edit must anchor to a UNIQUE location; expand the selection with surrounding context " +
+                "(still <=250 chars) so it appears exactly once.",
+            );
+          }
+          // Exactly one match: capture the RAW anchored substring to store as the
+          // comment selection (so apply-time equality holds). If this returns
+          // null despite countAnchorMatches===1 (shouldn't happen), fall back to
+          // the raw agent selection below rather than crash.
+          anchoredSelection = getAnchoredText(page.content, selection);
+          anchorNormalized = resolveAnchorSelection(
+            page.content,
+            selection,
+          ).normalized;
+        } else {
+          const resolved = resolveAnchorSelection(page.content, selection);
+          if (!resolved.found) {
+            throw this.anchorNotFoundError(page.content, selection, false);
+          }
+          anchorNormalized = resolved.normalized;
+        }
+      } catch (e) {
+        // Rethrow our own "not found"/"ambiguous"/"spans multiple blocks" errors;
+        // swallow read/network errors so the live anchor step can still try (and
+        // enforce) anchoring.
+        if (
+          e instanceof Error &&
+          (e.message.startsWith("createComment: could not find the selection") ||
+            e.message.startsWith(
+              "createComment: the selection spans multiple blocks",
+            ) ||
+            e.message.startsWith(
+              "createComment: the suggestion's selection is ambiguous",
+            ))
+        ) {
+          throw e;
+        }
+        if (process.env.DEBUG) {
+          console.error(
+            "Pre-check getPageJson failed; deferring to live anchor step:",
+            e,
+          );
+        }
+      }
+    }
+
+    // Convert through the full Docmost schema. Deliberately the NON-canonicalizing
+    // variant: a comment body may carry a footnote definition with no matching
+    // reference, and canonicalization would drop it (data loss). See
+    // markdownToProseMirror vs markdownToProseMirrorCanonical.
+    const jsonContent = await markdownToProseMirror(content);
+    const payload: Record<string, any> = {
+      pageId,
+      content: JSON.stringify(jsonContent),
+      type: effectiveType,
+    };
+    // For a suggestion, store the RAW anchored substring (anchoredSelection) so
+    // the stored selection === the text under the mark === apply-time
+    // expectedText. Ordinary comments (and the null fallback) keep the raw
+    // agent selection — their selection is only display/anchor and never used
+    // by apply, so their behavior is unchanged.
+    if (!isReply && selection)
+      payload.selection = anchoredSelection ?? selection;
+    if (parentCommentId) payload.parentCommentId = parentCommentId;
+    // Only a top-level inline comment (with a selection) may carry a suggestion.
+    if (!isReply && selection && hasSuggestion) {
+      payload.suggestedText = suggestedText;
+    }
+
+    const response = await this.client.post("/comments/create", payload);
+    const comment = response.data.data || response.data;
+    const markdown = comment.content
+      ? convertProseMirrorToMarkdown(this.parseCommentContent(comment.content))
+      : content;
+    const result: any = {
+      data: filterComment(comment, markdown),
+      success: true,
+    };
+
+    // Replies inherit the parent's anchor: no selection, no anchoring.
+    if (isReply) {
+      return result;
+    }
+
+    // Anchor the comment in the document. The /comments/create API records the
+    // comment + its `selection` text, but it does NOT insert the comment MARK
+    // into the page content, so without this the inline comment has no
+    // highlight/anchor and is not clickable. If anchoring fails the comment is
+    // rolled back (deleted) and an error is thrown — never an orphan comment.
+    const newCommentId: string = comment.id;
+    // Guard: a create response without an id would mean writing a comment mark
+    // with commentId: undefined and a later delete of a falsy id. We have no id
+    // to roll back here (nothing was created with an id), so just fail loudly.
+    if (!newCommentId) {
+      throw new Error(
+        "createComment: the server returned no comment id, so the comment could not be anchored",
+      );
+    }
+    let anchored = false;
+    // Set inside the transform when a suggestion's live anchor is ambiguous
+    // (>=2 occurrences), so the rollback path can surface the right error.
+    let ambiguousInLiveDoc = false;
+    // Captured inside the transform on a not-found abort, so the rollback path
+    // can surface the closest-block / spans-multiple-blocks hint built from the
+    // LIVE document (the pre-check page is not in scope there).
+    let liveNotFoundError: Error | null = null;
+    try {
+      const collabToken = await this.getCollabTokenWithReauth();
+      // Open the collab doc by the canonical UUID, never the slugId (#260). The
+      // /comments/create REST call above keeps the agent-supplied id.
+      const pageUuid = await this.resolvePageId(pageId);
+      // Route through the mutatePage seam (not the free function) so this
+      // wrapper's uniqueness gate + rollback can be unit-tested without a live
+      // Hocuspocus collab socket.
+      const mutation = await this.mutatePage(
+        pageUuid,
+        collabToken,
+        this.apiUrl,
+        (liveDoc) => {
+          const doc =
+            liveDoc && liveDoc.type === "doc"
+              ? liveDoc
+              : { type: "doc", content: [] };
+          if (hasSuggestion) {
+            // Authoritative uniqueness check against the LIVE document: a
+            // suggestion must anchor to EXACTLY ONE occurrence, otherwise
+            // "Apply" would rewrite the wrong/ambiguous text. If the live doc
+            // no longer has exactly one occurrence (it changed since the
+            // pre-check), abort so the just-created comment is rolled back
+            // rather than mis-anchored to the first occurrence.
+            const liveCount = countAnchorMatches(doc, selection as string);
+            if (liveCount !== 1) {
+              ambiguousInLiveDoc = liveCount >= 2;
+              if (liveCount === 0) {
+                liveNotFoundError = this.anchorNotFoundError(
+                  doc,
+                  selection as string,
+                  true,
+                );
+              }
+              return null;
+            }
+          }
+          if (applyAnchorInDoc(doc, selection as string, newCommentId)) {
+            anchored = true;
+            return doc;
+          }
+          // Selection text not found in the LIVE document: abort the write. The
+          // rollback + throw below turns this into a hard error.
+          liveNotFoundError = this.anchorNotFoundError(
+            doc,
+            selection as string,
+            true,
+          );
+          return null;
+        },
+      );
+      result.verify = mutation.verify;
+    } catch (e) {
+      // The comment record already exists; roll it back so we never leave an
+      // orphan, then rethrow the original anchoring error.
+      await this.safeDeleteComment(newCommentId);
+      throw e;
+    }
+
+    if (!anchored) {
+      // Mutation aborted because the selection was not found (or, for a
+      // suggestion, was ambiguous) in the live document. Roll back the comment
+      // and surface a hard error.
+      await this.safeDeleteComment(newCommentId);
+      if (ambiguousInLiveDoc) {
+        throw new Error(
+          "createComment: the suggestion's selection is ambiguous in the live document (multiple occurrences); the comment was rolled back. Expand the selection with surrounding context so it is unique.",
+        );
+      }
+      throw (
+        liveNotFoundError ??
+        new Error(
+          "createComment: failed to anchor the comment (selection not found in the live document); the comment was rolled back",
+        )
+      );
+    }
+
+    // Soft warning (like editPageText): the selection only matched after
+    // stripping markdown, so the caller likely quoted a styled fragment.
+    if (anchorNormalized) {
+      result.warning =
+        "The selection matched only after stripping markdown syntax; the comment " +
+        "was anchored on the document's plain text. Copy the selection verbatim " +
+        "from getPage / searchInPage output to avoid this.";
+    }
+
+    result.anchored = true;
+    return result;
+  }
+
+  /**
+   * Best-effort rollback of a just-created comment. Swallows any delete failure
+   * (logging under DEBUG) so a failed cleanup never masks the original error.
+   */
+  protected async safeDeleteComment(commentId: string): Promise<void> {
+    // Defense in depth: never call the delete API with a falsy id — there is
+    // nothing to roll back, and deleteComment(undefined) would hit a bad route.
+    if (!commentId) return;
+    try {
+      await this.deleteComment(commentId);
+    } catch (delErr) {
+      if (process.env.DEBUG) {
+        console.error(
+          "Failed to roll back comment after anchoring error:",
+          delErr,
+        );
+      }
+    }
+  }
+
+
+  async updateComment(commentId: string, content: string) {
+    // Fail fast (#436): reject a truncated id before any network call.
+    assertFullUuid("updateComment", "commentId", commentId);
+    await this.ensureAuthenticated();
+    // NON-canonicalizing on purpose (comment body — see createComment).
+    const jsonContent = await markdownToProseMirror(content);
+    await this.client.post("/comments/update", {
+      commentId,
+      content: JSON.stringify(jsonContent),
+    });
+    return {
+      success: true,
+      commentId,
+      message: "Comment updated successfully.",
+    };
+  }
+
+
+  async deleteComment(commentId: string) {
+    // Fail fast (#436): reject a truncated id before any network call.
+    assertFullUuid("deleteComment", "commentId", commentId);
+    await this.ensureAuthenticated();
+    return this.client
+      .post("/comments/delete", { commentId })
+      .then((res) => res.data);
+  }
+
+  /**
+   * Resolve or reopen a top-level comment thread (reversible — `resolved`
+   * toggles the state). Only top-level comments can be resolved; the server
+   * rejects resolving a reply. Hits POST /comments/resolve.
+   */
+  async resolveComment(commentId: string, resolved: boolean) {
+    // Fail fast (#436): reject a truncated id before any network call.
+    assertFullUuid("resolveComment", "commentId", commentId);
+    await this.ensureAuthenticated();
+    const response = await this.client.post("/comments/resolve", {
+      commentId,
+      resolved,
+    });
+    const comment = response.data?.data ?? response.data;
+    return {
+      success: true,
+      commentId,
+      resolved,
+      comment,
+    };
+  }
+
+  /**
+   * Check for new comments across pages in a space (optionally scoped to a
+   * subtree): pages updated after `since` are scanned and their comments
+   * filtered by createdAt > since.
+   */
+  async checkNewComments(
+    spaceId: string,
+    since: string,
+    parentPageId?: string,
+  ) {
+    await this.ensureAuthenticated();
+
+    const sinceDate = new Date(since);
+
+    // Reject an unparseable `since`: comparing against an Invalid Date silently
+    // yields zero new comments (every `>` against NaN is false), which would
+    // mask a malformed input as "nothing new" instead of erroring.
+    if (Number.isNaN(sinceDate.getTime())) {
+      throw new Error(
+        `checkNewComments: invalid "since" date "${since}"; expected an ISO-8601 timestamp`,
+      );
+    }
+
+    // 1. Enumerate the FULL set of pages in scope via the page tree (a complete
+    // page index), NOT the bounded "/pages/recent" feed which caps at ~5000
+    // recent items and silently misses comments on older pages.
+    //
+    // Subtree scope: when parentPageId is given, the scope is that page ITSELF
+    // plus every descendant. Otherwise the scope is the whole space (all roots
+    // and their descendants).
+    //
+    // NOTE: do NOT pre-filter by page.updatedAt — creating a comment does not
+    // bump it (verified on a live server), so such a filter silently misses
+    // comments on pages that were not otherwise edited. The complete tree walk
+    // already restricts the scope correctly, so no recent-feed allow-list is
+    // needed any more.
+    //
+    // The subtree scope (parentPageId given) already INCLUDES the root node
+    // itself: /pages/tree seeds getPageAndDescendants with id = parentPageId, so
+    // no separate getPageRaw fetch for the parent is needed.
+    const { pages: pagesInScope, truncated } = await this.enumerateSpacePages(
+      spaceId,
+      parentPageId,
+    );
+
+    // 2. Fetch comments for each page, keep ones created after since
+    const results: any[] = [];
+    for (const page of pagesInScope) {
+      try {
+        // Full feed (incl. resolved): a "new comments since" scan reports all
+        // recent activity; the active-only filter is scoped to listComments.
+        const comments = (await this.listComments(page.id, true)).items;
+        const newComments = comments.filter(
+          (c: any) => new Date(c.createdAt) > sinceDate,
+        );
+        if (newComments.length > 0) {
+          results.push({
+            pageId: page.id,
+            pageTitle: page.title,
+            comments: newComments,
+          });
+        }
+      } catch (e: any) {
+        // Skip pages with errors (e.g. deleted between calls)
+      }
+    }
+
+    const totalNewComments = results.reduce(
+      (sum, r) => sum + r.comments.length,
+      0,
+    );
+
+    // `truncated` is reported by enumerateSpacePages: it is true ONLY when the
+    // stdio fallback BFS hit its node cap. The primary /pages/tree path is
+    // uncapped, so a space with legitimately many pages is not falsely flagged.
+    return {
+      since,
+      scope: parentPageId ? `subtree of ${parentPageId}` : `space ${spaceId}`,
+      checkedPages: pagesInScope.length,
+      pagesWithNewComments: results.length,
+      totalNewComments,
+      truncated,
+      comments: results,
+    };
+  }
+
+  // --- Image upload / embedding ---
+
+  /** Map a Content-Type string to a supported MIME type, or null if unsupported. */
+  }
+  return CommentsMixin;
+}

packages/mcp/src/client/context.ts

@@ -0,0 +1,698 @@
+// Shared client context + core seams (issue #450). The abstract base of the
+// DocmostClient mixin chain: it owns ALL shared instance state (the axios
+// client, apiUrl, auth tokens, the resolvePageId cache, the collab-token cache,
+// the sandbox/metrics sinks) and the core HTTP/auth/pagination/write seams every
+// domain module builds on. Domain modules are mixins layered on top; the final
+// DocmostClient (client.ts) assembles them. Extracted VERBATIM from the original
+// monolith — only field/seam visibility was widened from `private` to
+// `protected` so sibling mixins can reach the shared state through `this`, and
+// the cross-module methods that live in other mixins are declared `abstract`
+// here so `this.<method>` type-checks. No behaviour changed.
+import axios, { AxiosInstance } from "axios";
+import FormData from "form-data";
+import {
+  updatePageContentRealtime,
+  replacePageContent,
+  markdownToProseMirror,
+  markdownToProseMirrorCanonical,
+  mutatePageContent,
+  assertYjsEncodable,
+  MutationResult,
+} from "../lib/collaboration.js";
+import { acquireCollabSession } from "../lib/collab-session.js";
+import { withPageLock, isUuid } from "../lib/page-lock.js";
+import { getCollabToken, performLogin } from "../lib/auth-utils.js";
+import { formatDocmostAxiosError } from "./errors.js";
+import { GetPageConversionCache } from "./getpage-cache.js";
+
+// A generic mixin base constructor (issue #450). Each domain mixin is a factory
+// `<T extends GConstructor<DocmostClientContext>>(Base: T) => class extends Base`
+// so the mixins compose into one prototype chain sharing this context.
+export type GConstructor<T = {}> = abstract new (...args: any[]) => T;
+
+/**
+ * Configuration for a DocmostClient / MCP server instance. A discriminated
+ * union: either service-account credentials (email/password — the client calls
+ * performLogin, powering the external /mcp HTTP endpoint and the stdio CLI) OR
+ * a token getter (getToken — the client uses the returned BARE access JWT as
+ * the Bearer and never calls performLogin; used for the internal per-user path).
+ *
+ * Both branches may ALSO carry an optional `getCollabToken` provider. When set,
+ * content mutations (which go over the collaboration websocket) use the token it
+ * returns INSTEAD of calling `POST /auth/collab-token`. The internal per-user
+ * agent path uses this to hand the client a provenance collab token (signed
+ * `actor:'agent'`+`aiChatId`), so agent content edits are attributed without a
+ * spoofable client-side field. When absent the client keeps the original
+ * `/auth/collab-token` path (service-account/stdio unchanged).
+ *
+ * Housed here (not in index.ts) so client.ts has no type dependency on index.ts;
+ * index.ts re-exports it for the package's public surface.
+ */
+// Sink the stash tool writes blobs into. The host app binds this to its in-RAM
+// SandboxStore and composes the public `uri` (the package never sees the store
+// or any env). `put` returns the anonymous read URL plus integrity metadata.
+export type SandboxPut = (
+  buf: Buffer,
+  mime: string,
+) => { uri: string; sha256: string; size: number };
+
+export type DocmostMcpConfig = { apiUrl: string } & (
+  | { email: string; password: string }
+  | { getToken: () => Promise<string> } // returns a BARE JWT; the client adds "Bearer "
+) & {
+    // Optional collab-token provider (returns a ready collab JWT). Common to
+    // both branches; see the type doc above.
+    getCollabToken?: () => Promise<string>;
+    // Optional blob sandbox sink. Present only where the stash tool is wired;
+    // when absent, stashPage throws a clear "not configured" error. The
+    // optional `has`/`evict` probes let stashPage keep its mirror counts honest
+    // under the store's FIFO eviction (see stashPage); older sinks omit them.
+    sandbox?: {
+      put: SandboxPut;
+      has?: (uri: string) => boolean;
+      evict?: (uri: string) => void;
+    };
+    // Dependency-neutral metrics sink. When present, the client emits generic
+    // (name, value, labels) samples; the HOST maps those names onto its own
+    // metrics registry (the package never depends on prom-client or the server).
+    // Absent in standalone/stdio mode → the client is a complete no-op here.
+    onMetric?: (
+      name: string,
+      value: number,
+      labels?: Record<string, string>,
+    ) => void;
+  };
+
+
+/**
+ * Collab-token cache TTL in milliseconds (issue #435). Read fresh from the
+ * environment on every mint — like collab-session.ts readConfig — so tests and a
+ * live rollback can change it without reloading the module.
+ *
+ * Why a cache at all: the live CollabSession registry (#400/#431) keys sessions
+ * on (wsUrl, pageId, collabToken) for identity isolation (invariant 4). But BOTH
+ * collab-token sources mint a FRESH token per mutation — the in-app provider
+ * re-signs a JWT whose iat/exp (seconds) changes every second, and the external
+ * MCP POSTs /auth/collab-token each call — so the token in the key changed on
+ * every op and the session was almost never reused (connect-storms, 25s
+ * timeouts, zombie sessions). Caching the token per-client keeps the key stable
+ * across a burst of mutations so ONE session is reused.
+ *
+ * Default 5 min: well under the 24h collab-token lifetime AND <= the collab
+ * session max-age (10 min, MCP_COLLAB_SESSION_MAX_AGE_MS), so the
+ * permission-staleness window is not widened beyond what #431 already accepted.
+ * The rollback knob is an EXPLICIT 0 (or a negative number): that DISABLES the
+ * cache — an exact fetch-per-call legacy path, mirroring how idleMs<=0 disables
+ * the session cache. Unset OR unparseable (e.g. a typo like "5min", "abc") falls
+ * back to the 5-min default with the cache ON — parseInt yields NaN, which is
+ * treated as "not configured", not as "disabled". So to turn the cache off you
+ * must set the value to exactly 0, not to garbage.
+ */
+function readCollabTokenTtlMs(): number {
+  const raw = parseInt(process.env.MCP_COLLAB_TOKEN_TTL_MS ?? "", 10);
+  return Number.isFinite(raw) ? Math.max(0, raw) : 5 * 60 * 1000;
+}
+
+export abstract class DocmostClientContext {
+  protected client: AxiosInstance;
+  protected token: string | null = null;
+  protected apiUrl: string;
+  // email/password are only set on the service-account (credentials) variant;
+  // null on the getToken variant (where there are no credentials to log in with).
+  protected email: string | null = null;
+  protected password: string | null = null;
+  // Per-user token provider. When set, login() calls it to obtain a BARE access
+  // JWT instead of performLogin, and the 401/403 re-auth path re-calls it.
+  protected getTokenFn: (() => Promise<string>) | null = null;
+  // Optional collab-token provider. When set, getCollabTokenWithReauth() returns
+  // its token instead of calling POST /auth/collab-token; on a 401/403 it is
+  // re-invoked once. Used by the internal agent to carry signed provenance.
+  protected getCollabTokenFn: (() => Promise<string>) | null = null;
+  // Optional blob-sandbox sink for the stash tool. Null when not configured.
+  protected sandboxPut: SandboxPut | null = null;
+  // Optional probes paired with the sink. `has` lets stashPage detect a blob
+  // FIFO-evicted by a LATER put in the same stash; `evict` lets it free this
+  // op's image blobs if the final doc put throws. Null when the sink omits them.
+  protected sandboxHas: ((uri: string) => boolean) | null = null;
+  protected sandboxEvict: ((uri: string) => void) | null = null;
+  // Optional dependency-neutral metrics sink (see DocmostMcpConfig.onMetric).
+  // Null on the legacy positional form and whenever the host omits it → no-op.
+  protected onMetricFn:
+    | ((name: string, value: number, labels?: Record<string, string>) => void)
+    | null = null;
+  // In-flight login dedup: when the token expires, the 401 interceptor,
+  // ensureAuthenticated, getCollabTokenWithReauth and the two multipart retries
+  // can all call login() at once. Memoizing a single promise collapses that
+  // thundering herd into ONE /auth/login request that everyone awaits.
+  protected loginPromise: Promise<void> | null = null;
+  // Canonical-UUID cache for resolvePageId: maps an agent-supplied slugId to the
+  // page's canonical UUID, so repeated collab edits on the same page do not
+  // re-fetch /pages/info. A UUID input short-circuits before this cache (see
+  // resolvePageId), so only slugId->uuid entries are stored/read here.
+  protected pageIdCache = new Map<string, string>();
+
+  // Collab-token cache (issue #435): the last minted collab token plus the
+  // wall-clock time it was minted, so a burst of content mutations reuses ONE
+  // token and therefore ONE live CollabSession (whose registry key includes the
+  // token — #400 invariant 4). Per-instance: a DocmostClient is built per
+  // user/per chat request, so a cached token can never leak across identities.
+  // Reset whenever the client's identity changes (login() / this.token cleared);
+  // bypassed on a forced refresh (the 401/403 reauth path). null = no token yet.
+  protected collabTokenCache: { token: string; mintedAt: number } | null = null;
+
+  // Content-addressed conversion cache for getPage (issue #479). Keyed on
+  // (canonical pageId, updatedAt, optionsHash) -> the converted Markdown, so a
+  // re-read of an UNCHANGED page skips the expensive convertProseMirrorToMarkdown
+  // tree walk. Per-instance (a DocmostClient is built per user / per chat), so a
+  // cached conversion can never leak across identities. See getpage-cache.ts.
+  protected getPageCache = new GetPageConversionCache();
+
+  // Two construction forms:
+  //  - new DocmostClient(config)                  // discriminated union (current)
+  //  - new DocmostClient(baseURL, email, password) // legacy positional creds
+  // The positional form is retained so existing callers/tests keep working; it
+  // is exactly equivalent to the credentials branch of the object form.
+  constructor(config: DocmostMcpConfig);
+  constructor(baseURL: string, email: string, password: string);
+  constructor(
+    configOrBaseURL: DocmostMcpConfig | string,
+    email?: string,
+    password?: string,
+  ) {
+    // Normalize the legacy positional form into the object union.
+    const config: DocmostMcpConfig =
+      typeof configOrBaseURL === "string"
+        ? { apiUrl: configOrBaseURL, email: email!, password: password! }
+        : configOrBaseURL;
+
+    this.apiUrl = config.apiUrl;
+    if ("getToken" in config) {
+      // Token variant: carry the user's JWT via getToken; no credentials, so
+      // login() must never call performLogin (there is nothing to log in with).
+      this.getTokenFn = config.getToken;
+    } else {
+      // Service-account variant: behaves exactly as before (performLogin).
+      this.email = config.email;
+      this.password = config.password;
+    }
+    // Optional, available to both variants. When present, content mutations get
+    // their collab token from here instead of POST /auth/collab-token.
+    if (config.getCollabToken) {
+      this.getCollabTokenFn = config.getCollabToken;
+    }
+    if (config.sandbox) {
+      this.sandboxPut = config.sandbox.put;
+      this.sandboxHas = config.sandbox.has ?? null;
+      this.sandboxEvict = config.sandbox.evict ?? null;
+    }
+    // Legacy positional form carries no onMetric → null (complete no-op).
+    this.onMetricFn = config.onMetric ?? null;
+    this.client = axios.create({
+      baseURL: this.apiUrl,
+      // Default request timeout so a hung connection cannot wedge a per-page
+      // lock or block the server indefinitely. Multipart uploads override this
+      // with a longer per-request timeout.
+      timeout: 30000,
+      headers: {
+        "Content-Type": "application/json",
+      },
+    });
+
+    // Re-authenticate transparently on a 401/403 once: the JWT authToken can
+    // expire while the server is long-running, after which every cached-token
+    // request would otherwise fail until a manual restart. On such a response,
+    // clear the stale token, perform a fresh login, and replay the original
+    // request exactly once (guarded by config._retry to avoid infinite loops;
+    // the login request itself is never retried).
+    this.client.interceptors.response.use(
+      (response) => response,
+      async (error) => {
+        const config = error.config;
+        const status = error.response?.status;
+        const isAuthError = status === 401 || status === 403;
+        const isLoginRequest =
+          typeof config?.url === "string" && config.url.includes("/auth/login");
+
+        if (config && isAuthError && !config._retry && !isLoginRequest) {
+          config._retry = true;
+          // Drop the stale token + Authorization header before re-login. Also
+          // clear the collab-token cache (#435): a new identity/login must not
+          // keep serving a collab token minted under the old one.
+          this.token = null;
+          this.collabTokenCache = null;
+          delete this.client.defaults.headers.common["Authorization"];
+          try {
+            await this.login();
+          } catch (loginError) {
+            // Re-login failed: surface the original error to the caller.
+            return Promise.reject(error);
+          }
+          // Re-issue the original request with the freshly minted Bearer token.
+          // Read it from the default header that login() just set, not from
+          // this.token, to avoid a theoretical "Bearer null" if this.token was
+          // cleared between login() resolving and this point.
+          config.headers = config.headers || {};
+          config.headers["Authorization"] =
+            this.client.defaults.headers.common["Authorization"];
+          return this.client.request(config);
+        }
+
+        return Promise.reject(error);
+      },
+    );
+
+    // Diagnostics interceptor (issue #437). Registered AFTER the re-login
+    // interceptor so a successful re-login retry (which resolves to a real
+    // response) is never seen here as an error; only a genuine failure reaches
+    // this rejection handler. It reformats error.message IN PLACE (see
+    // formatDocmostAxiosError — kept as a mutation, not a custom Error class, so
+    // the surrounding axios.isAxiosError / error.response?.status / config._retry
+    // checks keep working) and re-rejects the SAME error. The _docmostFormatted
+    // flag makes a re-processed retry-failure a no-op.
+    this.client.interceptors.response.use(
+      (response) => response,
+      (error) => {
+        formatDocmostAxiosError(error);
+        return Promise.reject(error);
+      },
+    );
+  }
+
+
+  // --- Cross-module seams (issue #450) -----------------------------------
+  // A method in one domain mixin sometimes calls a PROTECTED method owned by
+  // another mixin (e.g. nodes-write -> validateDocUrls in doc-validate). Those
+  // callees are `protected`, so they cannot be surfaced through the public
+  // per-mixin interfaces. Declaring them here on the shared base lets `this.<m>`
+  // type-check across modules. Each is a stub that is ALWAYS overridden by the
+  // owning mixin (layered above this base in the chain), so the body never runs;
+  // it throws only to make an impossible mis-wiring loud instead of silent.
+  // (The PUBLIC cross-module callees — getPage, getPageJson, listComments,
+  // deleteComment, listPageHistory — arrive via the mixins' public interfaces,
+  // so they are not restated here.)
+  protected enumerateSpacePages(
+    _spaceId: string,
+    _rootPageId?: string,
+  ): Promise<{ pages: any[]; truncated: boolean }> {
+    throw new Error("enumerateSpacePages not wired (missing ReadMixin)");
+  }
+  protected validateDocUrls(_node: any, _depth?: number): void {
+    throw new Error("validateDocUrls not wired (missing DocValidateMixin)");
+  }
+  protected validateDocStructure(_node: any, _depth?: number): void {
+    throw new Error("validateDocStructure not wired (missing DocValidateMixin)");
+  }
+  protected assertValidNodeShape(_op: string, _node: any): void {
+    throw new Error("assertValidNodeShape not wired (missing DocValidateMixin)");
+  }
+  protected fetchInternalFile(
+    _src: string,
+  ): Promise<{ buffer: Buffer; mime: string }> {
+    throw new Error("fetchInternalFile not wired (missing StashMixin)");
+  }
+  protected uploadAttachmentBuffer(
+    _pageId: string,
+    _buffer: Buffer,
+    _fileName: string,
+    _mime: string,
+  ): Promise<{ id: string; fileName: string; fileSize: number }> {
+    throw new Error("uploadAttachmentBuffer not wired (missing MediaMixin)");
+  }
+  protected fetchAttachmentText(_src: string): Promise<string> {
+    throw new Error("fetchAttachmentText not wired (missing MediaMixin)");
+  }
+  // PUBLIC cross-module callees. Declared here too (as always-overridden stubs)
+  // so a mixin calling e.g. `this.getPageJson` type-checks against the base —
+  // the mixin's own public interface only covers its own methods. The real
+  // implementations live in ReadMixin / CommentsMixin / PagesMixin and shadow
+  // these on the prototype chain.
+  getPage(_pageId: string): Promise<any> {
+    throw new Error("getPage not wired (missing ReadMixin)");
+  }
+  getPageJson(_pageId: string): Promise<any> {
+    throw new Error("getPageJson not wired (missing ReadMixin)");
+  }
+  listComments(_pageId: string, _includeResolved?: boolean): Promise<any> {
+    throw new Error("listComments not wired (missing CommentsMixin)");
+  }
+  deleteComment(_commentId: string): Promise<any> {
+    throw new Error("deleteComment not wired (missing CommentsMixin)");
+  }
+  listPageHistory(_pageId: string, _cursor?: string): Promise<any> {
+    throw new Error("listPageHistory not wired (missing PagesMixin)");
+  }
+
+  /** Application base URL (API URL without the /api suffix). */
+  get appUrl(): string {
+    return this.apiUrl.replace(/\/api\/?$/, "");
+  }
+
+
+  async login() {
+    // Reuse an in-flight login if one is already running so concurrent callers
+    // share a single token fetch instead of each issuing their own.
+    if (!this.loginPromise) {
+      // Token variant: re-fetch a BARE JWT via getToken() (there are no
+      // credentials to log in with — on a 401/403 the interceptor below calls
+      // login() again, which re-invokes getToken()). Credentials variant:
+      // performLogin against /auth/login exactly as before.
+      const fetchToken = this.getTokenFn
+        ? this.getTokenFn()
+        : performLogin(this.apiUrl, this.email!, this.password!);
+      this.loginPromise = fetchToken
+        .then((token) => {
+          // Guard against an empty/invalid token (e.g. a getToken provider that
+          // resolves to "" or null): without this an empty token would set a
+          // literal "Authorization: Bearer null"/"Bearer " header and every
+          // request would 401 with a confusing error. Fail loudly instead.
+          if (typeof token !== "string" || token.length === 0) {
+            throw new Error("getToken returned an empty token");
+          }
+          this.token = token;
+          // Identity (re)established: drop any collab token minted under a
+          // previous identity so the #435 cache can never outlive it.
+          this.collabTokenCache = null;
+          this.client.defaults.headers.common["Authorization"] =
+            `Bearer ${token}`;
+        })
+        .finally(() => {
+          this.loginPromise = null;
+        });
+    }
+    return this.loginPromise;
+  }
+
+
+  async ensureAuthenticated() {
+    if (!this.token) {
+      await this.login();
+    }
+  }
+
+  /**
+   * Fetch a collaboration token, transparently re-authenticating once on a
+   * 401/403. getCollabToken() uses bare axios internally, so it is NOT covered
+   * by this.client's response interceptor; this helper replicates that
+   * behaviour for collab-token requests: ensure a token, try once, and on an
+   * expired-token auth error perform a fresh login and retry exactly once.
+   *
+   * Collab-token cache (issue #435): both sources — the getCollabToken provider
+   * (in-app agent) AND the REST /auth/collab-token endpoint (external MCP) — mint
+   * a FRESH token per call, whose string therefore changes every op. Since the
+   * live CollabSession registry keys on the token string (#400/#431 invariant 4),
+   * that churned the key and defeated session reuse. So we cache the last minted
+   * token per-client for readCollabTokenTtlMs() and hand it back for a burst of
+   * mutations, keeping the session key stable. `forceRefresh` bypasses the cache
+   * (the 401/403 reauth retry uses it, so the retry cannot be handed the same
+   * stale token that just failed — otherwise reauth would be a no-op). TTL 0
+   * disables the cache: exact fetch-per-call legacy behaviour.
+   */
+  protected async getCollabTokenWithReauth(
+    forceRefresh = false,
+  ): Promise<string> {
+    const ttl = readCollabTokenTtlMs();
+    // Serve the cached collab token while it is still fresh (identity isolation
+    // is preserved: the cache is a per-instance field on a client built per
+    // user/per chat request, and it is cleared on every identity change).
+    if (
+      !forceRefresh &&
+      ttl > 0 &&
+      this.collabTokenCache &&
+      Date.now() - this.collabTokenCache.mintedAt < ttl
+    ) {
+      return this.collabTokenCache.token;
+    }
+
+    // Collab-token PROVIDER path: when a getCollabToken provider was supplied
+    // (the internal agent's provenance collab token), use it instead of the
+    // REST /auth/collab-token endpoint. Re-invoke it once on a 401/403 (e.g. the
+    // signed token expired between content mutations in a long agent turn).
+    if (this.getCollabTokenFn) {
+      try {
+        const token = await this.getCollabTokenFn();
+        if (typeof token !== "string" || token.length === 0) {
+          throw new Error("getCollabToken returned an empty token");
+        }
+        return this.rememberCollabToken(token, ttl);
+      } catch (e) {
+        // On an auth error retry EXACTLY once, forcing a refresh so the retry
+        // re-invokes the provider (bypassing the cache) for a genuinely fresh
+        // token. `!forceRefresh` bounds it to a single retry (no loop).
+        if (this.isCollabAuthError(e) && !forceRefresh) {
+          return this.getCollabTokenWithReauth(true);
+        }
+        throw e;
+      }
+    }
+
+    await this.ensureAuthenticated();
+    try {
+      const token = await getCollabToken(this.apiUrl, this.token!);
+      return this.rememberCollabToken(token, ttl);
+    } catch (e) {
+      // getCollabToken wraps the AxiosError in a plain Error but attaches the
+      // HTTP status as `.status`, so isCollabAuthError detects an auth failure
+      // via either the raw AxiosError shape OR the attached status.
+      if (this.isCollabAuthError(e) && !forceRefresh) {
+        // Fresh login (which clears this.token AND the collab-token cache), then
+        // retry exactly once with the cache bypassed via forceRefresh.
+        await this.login();
+        return this.getCollabTokenWithReauth(true);
+      }
+      throw e;
+    }
+  }
+
+  /**
+   * Store a freshly minted collab token in the per-client cache (issue #435) and
+   * return it unchanged. No-op write when the cache is disabled (ttl<=0) or the
+   * token is empty, so a disabled cache is exact fetch-per-call legacy behaviour
+   * and a bad token is never cached.
+   */
+  protected rememberCollabToken(token: string, ttl: number): string {
+    if (ttl > 0 && typeof token === "string" && token.length > 0) {
+      this.collabTokenCache = { token, mintedAt: Date.now() };
+    }
+    return token;
+  }
+
+  /**
+   * True when an error carries a 401/403 — either as a raw AxiosError
+   * (`error.response.status`) or as the plain-Error `.status` that
+   * lib/auth-utils.getCollabToken attaches after wrapping the AxiosError.
+   */
+  protected isCollabAuthError(e: unknown): boolean {
+    const axiosStatus = axios.isAxiosError(e) ? e.response?.status : undefined;
+    const attachedStatus = (e as any)?.status;
+    return (
+      axiosStatus === 401 ||
+      axiosStatus === 403 ||
+      attachedStatus === 401 ||
+      attachedStatus === 403
+    );
+  }
+
+  /**
+   * Connect to the collaboration websocket, read the live doc, apply
+   * `transform`, write the result, and wait for the server to persist it —
+   * WITHOUT acquiring the per-page lock.
+   *
+   * This mirrors collaboration.mutatePageContent EXCEPT that it does not call
+   * withPageLock. It exists solely so replaceImage can hold ONE withPageLock
+   * across its scan -> upload -> write sequence: the per-page mutex is NOT
+   * reentrant, so calling the normal (self-locking) mutatePageContent inside an
+   * outer withPageLock for the same pageId would deadlock. The caller MUST hold
+   * the page lock for the whole operation; this helper assumes that invariant.
+   *
+   * `transform` receives the live ProseMirror doc and returns the NEW full doc
+   * to write, or `null` to abort with no write. Errors thrown by `transform`
+   * propagate to the caller.
+   *
+   * Resolves a `MutationResult { doc, verify }` mirroring mutatePageContent, so
+   * every content mutator (including replaceImage) can return a verifiable
+   * change report. The report is computed AFTER the atomic read->write and
+   * never throws.
+   */
+  protected async mutateLiveContentUnlocked(
+    pageId: string,
+    collabToken: string,
+    transform: (liveDoc: any) => any | null,
+  ): Promise<MutationResult> {
+    // Reuse a live CollabSession for the page (issue #400) instead of opening a
+    // fresh provider per op. acquireCollabSession does NOT take the per-page
+    // lock — the caller (replaceImage) already holds ONE withPageLock across its
+    // scan -> upload -> write sequence, and the mutex is not reentrant, so
+    // taking it here would deadlock. The synchronous read->write section and the
+    // unsyncedChanges/connectionLost ack logic live in CollabSession.mutate,
+    // preserved verbatim from the old inline machine (incl. the #152 structural
+    // diff that keeps a live editor's cursor anchored).
+    const session = await acquireCollabSession(pageId, collabToken, this.apiUrl, {
+      // Only the actual 25s collab connect timeout emits this — the connect-vs-
+      // unload signal; the other failure paths must NOT emit it.
+      onConnectTimeout: () =>
+        this.onMetricFn?.("collab_connect_timeouts_total", 1),
+    });
+    try {
+      return await session.mutate(transform);
+    } catch (e) {
+      // Drop the session on any failure so the next call reconnects fresh.
+      session.destroy("mutate failed");
+      throw e;
+    }
+  }
+
+  /**
+   * Generic pagination handler for Docmost API endpoints
+   */
+  async paginateAll<T = any>(
+    endpoint: string,
+    basePayload: Record<string, any> = {},
+    limit: number = 100,
+  ): Promise<T[]> {
+    await this.ensureAuthenticated();
+
+    const clampedLimit = Math.max(1, Math.min(100, limit));
+
+    // Hard ceiling on the number of pages to fetch: guards against a server
+    // that returns a perpetually-true hasNextPage (which would otherwise loop
+    // forever and accumulate duplicates).
+    const MAX_PAGES = 50;
+
+    let cursor: string | undefined;
+    let allItems: T[] = [];
+    let truncated = false;
+
+    for (let page = 0; page < MAX_PAGES; page++) {
+      const payload: Record<string, any> = {
+        ...basePayload,
+        limit: clampedLimit,
+      };
+      if (cursor) payload.cursor = cursor;
+
+      const response = await this.client.post(endpoint, payload);
+
+      const data = response.data;
+      const items = data.data?.items || data.items || [];
+      const meta = data.data?.meta || data.meta;
+
+      allItems = allItems.concat(items);
+
+      // Advance strictly via the server-issued cursor. A missing nextCursor (or
+      // hasNextPage false) means we reached the end. A cursor identical to the
+      // one we just sent means the server did not understand our pagination
+      // param — stop instead of re-fetching page one forever and duplicating.
+      const next = meta?.hasNextPage ? meta?.nextCursor : null;
+      if (!next || next === cursor) {
+        // If the server still reports more pages but stopped issuing a usable
+        // cursor at the ceiling, flag the result as truncated below.
+        if (page === MAX_PAGES - 1 && meta?.hasNextPage) truncated = true;
+        break;
+      }
+      cursor = next;
+
+      // Reaching the ceiling with more pages still available means the result
+      // set is truncated.
+      if (page === MAX_PAGES - 1) truncated = true;
+    }
+
+    // If the loop stopped because it hit the MAX_PAGES ceiling while the server
+    // still reported more results, the result set is truncated — warn so the
+    // caller is not silently handed an incomplete list.
+    if (truncated) {
+      console.warn(
+        `paginateAll: results from "${endpoint}" truncated at the ${MAX_PAGES}-page cap; more pages exist on the server`,
+      );
+    }
+
+    return allItems;
+  }
+
+
+  /** Raw page info including the ProseMirror JSON content and slugId. */
+  async getPageRaw(pageId: string) {
+    await this.ensureAuthenticated();
+    const response = await this.client.post("/pages/info", { pageId });
+    return response.data?.data ?? response.data;
+  }
+
+  /**
+   * Resolve an agent-supplied pageId to the page's CANONICAL UUID (`page.id`),
+   * so every collaboration document the MCP opens is named `page.<uuid>` — the
+   * SAME name the web editor always uses (`page.${page.id}`).
+   *
+   * The agent commonly passes a 10-char public slugId (from URLs/listings) as
+   * the pageId. The web editor opens the collab doc by UUID, but the MCP used to
+   * pass that slugId straight into the collab doc name (`page.<slugId>`). For one
+   * DB row that produced TWO independent Yjs documents whose debounced stores
+   * clobbered each other — the agent's edit was silently lost (#260).
+   *
+   * A UUID input short-circuits with no network round-trip. A slugId is resolved
+   * once via getPageRaw and cached (both slugId->uuid and uuid->uuid), so
+   * repeated edits on the same page add no extra request.
+   */
+  protected async resolvePageId(pageId: string): Promise<string> {
+    if (isUuid(pageId)) return pageId;
+    const cached = this.pageIdCache.get(pageId);
+    if (cached) return cached;
+    const data = await this.getPageRaw(pageId);
+    const uuid = data?.id;
+    if (typeof uuid !== "string" || !uuid) {
+      throw new Error(
+        `Could not resolve a canonical page id for "${pageId}"`,
+      );
+    }
+    this.pageIdCache.set(pageId, uuid);
+    return uuid;
+  }
+
+
+  /**
+   * Page-locked write seam over collaboration.mutatePageContent. Production just
+   * delegates; it exists as an overridable method so the insertFootnote wrapper
+   * (transform abort-on-not-found + response shaping) can be unit-tested without
+   * standing up a live Hocuspocus collab socket.
+   *
+   * SELF-RESOLVES the pageId to the canonical UUID (issue #449, "resolve-then-
+   * lock"): every write must lock and key its CollabSession by the UUID, never a
+   * raw slugId (#260). resolvePageId is cached/idempotent, so a caller that
+   * already resolved pays no extra round-trip; centralizing it here means a
+   * caller that reaches this seam with a raw slugId still locks correctly instead
+   * of silently splitting the mutex key. withPageLock also asserts the key is a
+   * UUID as a hard backstop.
+   */
+  protected async mutatePage(
+    pageId: string,
+    collabToken: string,
+    apiUrl: string,
+    transform: (doc: any) => any,
+  ): Promise<{ doc?: any; verify?: any }> {
+    const pageUuid = await this.resolvePageId(pageId);
+    return mutatePageContent(pageUuid, collabToken, apiUrl, transform);
+  }
+
+  /**
+   * Full-document write seam over collaboration.replacePageContent. Production
+   * just delegates; it exists as an overridable method so the full-doc write
+   * tools (updatePageJson, copyPageContent) can have their footnote-
+   * canonicalization binding unit-tested without a live Hocuspocus collab socket.
+   *
+   * SELF-RESOLVES the pageId to the canonical UUID (issue #449, "resolve-then-
+   * lock") for the same reason as mutatePage above — the lock/CollabSession key
+   * is guaranteed canonical here, not left to the caller's discipline.
+   */
+  protected async replacePage(
+    pageId: string,
+    doc: any,
+    collabToken: string,
+    apiUrl: string,
+  ): Promise<{ doc?: any; verify?: any }> {
+    const pageUuid = await this.resolvePageId(pageId);
+    return replacePageContent(pageUuid, doc, collabToken, apiUrl);
+  }
+
+  /**
+   * Export a page to a single self-contained Docmost-flavoured markdown file:
+   * meta block + body (with inline comment anchors + diagrams) + comment
+   * threads. Lossless round-trip target; see importPageMarkdown for the inverse.
+   */
+}

packages/mcp/src/client/doc-validate.ts

@@ -0,0 +1,248 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import {
+  updatePageContentRealtime,
+  replacePageContent,
+  markdownToProseMirror,
+  markdownToProseMirrorCanonical,
+  mutatePageContent,
+  assertYjsEncodable,
+  MutationResult,
+} from "../lib/collaboration.js";
+import {
+  replaceNodeById,
+  replaceNodeByIdWithMany,
+  reassignCollidingBlockIds,
+  deleteNodeById,
+  assertUnambiguousMatch,
+  insertNodeRelative,
+  insertNodesRelative,
+  blockPlainText,
+  buildOutline,
+  getNodeByRef,
+  readTable,
+  insertTableRow,
+  deleteTableRow,
+  updateTableCell,
+  findInvalidNode,
+} from "@docmost/prosemirror-markdown";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+
+// Public method surface of DocValidateMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements IDocValidateMixin` fails to compile on drift.
+export interface IDocValidateMixin {
+}
+
+export function DocValidateMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & IDocValidateMixin> & TBase {
+  abstract class DocValidateMixin extends Base implements IDocValidateMixin {
+  /**
+   * Validate a URL string against a scheme allowlist for a given context.
+   *
+   * The markdown link path enforces safe schemes via TipTap, but the raw
+   * JSON path (updatePageJson) bypasses that — so this is the sanitization
+   * choke point for ProseMirror JSON written directly by the caller.
+   *
+   * - "link":  reject javascript:, vbscript:, data: (any scheme that can
+   *            execute or smuggle script when the href is clicked).
+   * - "src":   allow only http(s):, mailto:, /api/files paths, or a
+   *            scheme-less relative/absolute path; reject
+   *            javascript:/vbscript:/data:/file:.
+   */
+  protected isSafeUrl(url: unknown, context: "link" | "src"): boolean {
+    if (typeof url !== "string") return false;
+    const trimmed = url.trim();
+    if (trimmed === "") return true; // empty href/src is harmless
+
+    // Extract a leading "scheme:" if present. A scheme must start with a
+    // letter and contain only letters/digits/+/-/. before the colon. Strip
+    // whitespace and ASCII control chars first so a tab/newline embedded in
+    // the scheme cannot smuggle a dangerous scheme past the check.
+    const cleaned = trimmed.replace(/[\s\x00-\x1f]+/g, "");
+    const schemeMatch = /^([a-zA-Z][a-zA-Z0-9+.-]*):/.exec(cleaned);
+    const scheme = schemeMatch ? schemeMatch[1].toLowerCase() : null;
+
+    const dangerous = new Set(["javascript", "vbscript", "data", "file"]);
+
+    if (context === "link") {
+      if (scheme === null) return true; // relative/anchor link is fine
+      // For links, data: is also blocked (can carry script payloads).
+      return !new Set(["javascript", "vbscript", "data"]).has(scheme);
+    }
+
+    // context === "src"
+    if (scheme === null) return true; // relative/absolute path (incl. /api/files)
+    if (dangerous.has(scheme)) return false;
+    return scheme === "http" || scheme === "https" || scheme === "mailto";
+  }
+
+  /**
+   * Recursively walk a ProseMirror doc and reject any unsafe URL on a link
+   * mark href or on a media node's src/url. Media nodes covered: image,
+   * attachment, video, plus embed (rendered as an iframe), youtube, drawio
+   * and excalidraw — all of which carry a user-controlled URL that Docmost
+   * renders. Throws a clear error on the first violation. A max-depth guard
+   * turns an over-deep document into a clean error instead of a RangeError
+   * stack overflow.
+   */
+  protected validateDocUrls(node: any, depth: number = 0): void {
+    const MAX_DEPTH = 200;
+    if (depth > MAX_DEPTH) {
+      throw new Error(
+        `document nesting exceeds the maximum depth of ${MAX_DEPTH}`,
+      );
+    }
+    if (!node || typeof node !== "object") return;
+
+    // Link marks on text nodes: validate the href.
+    if (Array.isArray(node.marks)) {
+      for (const mark of node.marks) {
+        if (mark && mark.type === "link" && mark.attrs) {
+          if (!this.isSafeUrl(mark.attrs.href, "link")) {
+            throw new Error(`unsafe link href rejected: "${mark.attrs.href}"`);
+          }
+        }
+      }
+    }
+
+    // Media nodes: validate src/url against the stricter src allowlist.
+    // embed renders as an iframe (highest risk); youtube/drawio/excalidraw
+    // likewise carry a user-controlled URL Docmost renders, so they get the
+    // same scheme check as image/attachment/video.
+    if (
+      node.type === "image" ||
+      node.type === "attachment" ||
+      node.type === "video" ||
+      node.type === "embed" ||
+      node.type === "youtube" ||
+      node.type === "drawio" ||
+      node.type === "excalidraw" ||
+      node.type === "audio" ||
+      node.type === "pdf"
+    ) {
+      const attrs = node.attrs || {};
+      for (const key of ["src", "url"]) {
+        if (attrs[key] != null && !this.isSafeUrl(attrs[key], "src")) {
+          throw new Error(
+            `unsafe ${node.type} ${key} rejected: "${attrs[key]}"`,
+          );
+        }
+      }
+    }
+
+    if (Array.isArray(node.content)) {
+      for (const child of node.content) {
+        this.validateDocUrls(child, depth + 1);
+      }
+    }
+  }
+
+  /**
+   * Recursively validate the STRUCTURE of a ProseMirror node (reuses the
+   * recursion shape of validateDocUrls). Every node must be an object with a
+   * string `type`; when present, `content` must be an array, `marks` must be
+   * an array of objects each with a string `type`, and a text node's `text`
+   * must be a string. Throws a clear "invalid ProseMirror document" error on
+   * the first violation. A max-depth guard turns an over-deep document into a
+   * clean error instead of a RangeError stack overflow.
+   */
+  protected validateDocStructure(node: any, depth: number = 0): void {
+    const MAX_DEPTH = 200;
+    if (depth > MAX_DEPTH) {
+      throw new Error(
+        `invalid ProseMirror document: nesting exceeds the maximum depth of ${MAX_DEPTH}`,
+      );
+    }
+    if (!node || typeof node !== "object" || typeof node.type !== "string") {
+      throw new Error(
+        "invalid ProseMirror document: every node must be an object with a string `type`",
+      );
+    }
+    if (
+      "text" in node &&
+      node.type === "text" &&
+      typeof node.text !== "string"
+    ) {
+      throw new Error(
+        "invalid ProseMirror document: a text node must have a string `text`",
+      );
+    }
+    if (node.marks !== undefined) {
+      if (!Array.isArray(node.marks)) {
+        throw new Error(
+          "invalid ProseMirror document: `marks` must be an array",
+        );
+      }
+      for (const mark of node.marks) {
+        if (
+          !mark ||
+          typeof mark !== "object" ||
+          typeof mark.type !== "string"
+        ) {
+          throw new Error(
+            "invalid ProseMirror document: every mark must be an object with a string `type`",
+          );
+        }
+      }
+    }
+    if (node.content !== undefined) {
+      if (!Array.isArray(node.content)) {
+        throw new Error(
+          "invalid ProseMirror document: `content` must be an array when present",
+        );
+      }
+      for (const child of node.content) {
+        this.validateDocStructure(child, depth + 1);
+      }
+    }
+  }
+
+  /**
+   * Pre-write SHAPE gate (#409). Walk the WHOLE node tree with the shared
+   * `findInvalidNode` and throw a rich, path-anchored error the instant a nested
+   * node has an absent/unknown `type` (or an unknown mark) — the exact shape that
+   * otherwise surfaces DEEP in the Yjs encode as the cryptic
+   * `Unknown node type: undefined`, but only AFTER a collab session was opened
+   * and a page lock taken. Calling this BEFORE `getCollabTokenWithReauth` /
+   * `mutatePageContent` fails fast: no collab connection, no lock, deterministic
+   * message. `op` names the tool for the message prefix (e.g. "patchNode").
+   *
+   * `findInvalidNode` derives its "known type" set from the very same
+   * `docmostExtensions` the encode path uses, so a node this gate accepts is one
+   * the encoder will accept too.
+   */
+  protected assertValidNodeShape(op: string, node: any): void {
+    const bad = findInvalidNode(node);
+    if (bad) {
+      throw new Error(`${op}: invalid node — ${bad.summary}`);
+    }
+  }
+
+  /**
+   * Replace page content with a raw ProseMirror JSON document (lossless) and/or
+   * update its title. Both `doc` and `title` are optional, but at least one must
+   * be supplied:
+   *  - `doc` provided   -> validate + full-overwrite the body (and update the
+   *                        title too when `title` is also given).
+   *  - `doc` omitted, `title` given -> title-only update; the body is NOT
+   *                        touched/resent (no collab write happens).
+   *  - neither given    -> throws (nothing to update).
+   */
+  }
+  return DocValidateMixin;
+}

packages/mcp/src/client/drawio.ts

@@ -0,0 +1,707 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import { parseCells as parseDrawioCells } from "../lib/drawio-xml.js";
+import {
+  replaceNodeById,
+  replaceNodeByIdWithMany,
+  reassignCollidingBlockIds,
+  deleteNodeById,
+  assertUnambiguousMatch,
+  insertNodeRelative,
+  insertNodesRelative,
+  blockPlainText,
+  buildOutline,
+  getNodeByRef,
+  readTable,
+  insertTableRow,
+  deleteTableRow,
+  updateTableCell,
+  findInvalidNode,
+} from "@docmost/prosemirror-markdown";
+import {
+  prepareModel,
+  decodeDrawioSvg,
+  buildDrawioSvg,
+  mxHash,
+  normalizeXml,
+  countUserCells,
+} from "../lib/drawio-xml.js";
+import { renderDiagramShapes } from "../lib/drawio-preview.js";
+import { applyElkLayout } from "../lib/drawio-layout.js";
+import {
+  buildFromGraph,
+  type Graph,
+  type LayoutMode as GraphLayoutMode,
+} from "../lib/drawio-graph.js";
+import { applyCellOps, type CellOp } from "../lib/drawio-cell-ops.js";
+import { mermaidToGraph } from "../lib/drawio-mermaid.js";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+
+// Public method surface of DrawioMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements IDrawioMixin` fails to compile on drift.
+export interface IDrawioMixin {
+  drawioGet(pageId: string, node: string, format?: "xml" | "svg"): Promise<{ pageId: string; nodeId: string; format: "xml" | "svg"; content: string; meta: { attachmentId: string | null; title: string | null; width: number | null; height: number | null; cellCount: number; hash: string; }; }>;
+  drawioCreate(pageId: string, where: { position: "before" | "after" | "append"; anchorNodeId?: string; anchorText?: string; }, xml: string, title?: string, layout?: "elk"): Promise<{ success: boolean; nodeId: string; attachmentId: string; warnings: string[]; verify?: any; }>;
+  drawioUpdate(pageId: string, node: string, xml: string, baseHash: string, layout?: "elk"): Promise<{ success: boolean; nodeId: string; attachmentId: string; warnings: string[]; verify?: any; }>;
+  drawioEditCells(pageId: string, node: string, operations: CellOp[], baseHash: string): Promise<{ success: boolean; nodeId: string; attachmentId: string; warnings: string[]; verify?: any; }>;
+  drawioFromGraph(pageId: string, where: { position: "before" | "after" | "append"; anchorNodeId?: string; anchorText?: string; }, graph: Graph, direction?: "LR" | "RL" | "TB" | "BT", preset?: string, layout?: GraphLayoutMode, node?: string): Promise<{ success: boolean; nodeId: string; attachmentId: string; warnings: string[]; iconsResolved: number; iconsMissing: string[]; verify?: any; }>;
+  drawioFromMermaid(pageId: string, where: { position: "before" | "after" | "append"; anchorNodeId?: string; anchorText?: string; }, mermaid: string, preset?: string): Promise<{ success: boolean; nodeId: string; attachmentId: string; warnings: string[]; iconsResolved: number; iconsMissing: string[]; verify?: any; }>;
+}
+
+export function DrawioMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & IDrawioMixin> & TBase {
+  abstract class DrawioMixin extends Base implements IDrawioMixin {
+  /**
+   * Resolve a drawio node on a page by `attrs.id` or `#<index>` and return the
+   * node plus its ref. Throws a clear error if the ref does not resolve to a
+   * drawio node.
+   */
+  protected async resolveDrawioNode(
+    pageId: string,
+    node: string,
+  ): Promise<{ node: any; ref: string }> {
+    const data = await this.getPageRaw(pageId);
+    const hit = getNodeByRef(
+      data.content ?? { type: "doc", content: [] },
+      node,
+    );
+    if (!hit) {
+      throw new Error(
+        `drawio: no node found for "${node}" on page ${pageId} (use the drawio node's attrs.id or "#<index>" from getOutline)`,
+      );
+    }
+    if (hit.type !== "drawio") {
+      throw new Error(
+        `drawio: node "${node}" on page ${pageId} is a ${hit.type}, not a drawio diagram`,
+      );
+    }
+    return { node: hit.node, ref: node };
+  }
+
+  /**
+   * Read a drawio diagram as mxGraph XML (default) or as the raw `.drawio.svg`.
+   * Runs the decode chain (base64/entity content= → drawio file → nested XML or
+   * pako-inflated compressed <diagram>). The returned `hash` is the
+   * optimistic-lock key for drawioUpdate.
+   */
+  async drawioGet(
+    pageId: string,
+    node: string,
+    format: "xml" | "svg" = "xml",
+  ): Promise<{
+    pageId: string;
+    nodeId: string;
+    format: "xml" | "svg";
+    content: string;
+    meta: {
+      attachmentId: string | null;
+      title: string | null;
+      width: number | null;
+      height: number | null;
+      cellCount: number;
+      hash: string;
+    };
+  }> {
+    await this.ensureAuthenticated();
+    const { node: drawio } = await this.resolveDrawioNode(pageId, node);
+    const attrs = drawio.attrs || {};
+    const src = attrs.src;
+    if (!src) {
+      throw new Error(
+        `drawio: node "${node}" on page ${pageId} has no src to read`,
+      );
+    }
+    const svg = await this.fetchAttachmentText(src);
+    const modelXml = decodeDrawioSvg(svg);
+    const meta = {
+      attachmentId: attrs.attachmentId ?? null,
+      title: attrs.title ?? null,
+      width: attrs.width != null ? Number(attrs.width) : null,
+      height: attrs.height != null ? Number(attrs.height) : null,
+      cellCount: countUserCells(modelXml),
+      hash: mxHash(modelXml),
+    };
+    return {
+      pageId,
+      nodeId: attrs.id ?? node,
+      format,
+      content: format === "svg" ? svg : normalizeXml(modelXml),
+      meta,
+    };
+  }
+
+  /**
+   * Create a drawio diagram from mxGraph XML: lint → schematic SVG preview
+   * (pure TS) → build the `.drawio.svg` (createDrawioSvg contract) → create the
+   * attachment → insert a `drawio` node before/after an anchor or appended.
+   * `xml` is a bare `<mxGraphModel>` or a list of `<mxCell>` (the server wraps
+   * it and adds the id=0/id=1 sentinels).
+   */
+  async drawioCreate(
+    pageId: string,
+    where: {
+      position: "before" | "after" | "append";
+      anchorNodeId?: string;
+      anchorText?: string;
+    },
+    xml: string,
+    title?: string,
+    layout?: "elk",
+  ): Promise<{
+    success: boolean;
+    nodeId: string;
+    attachmentId: string;
+    warnings: string[];
+    verify?: any;
+  }> {
+    await this.ensureAuthenticated();
+    if (
+      !where ||
+      (where.position !== "before" &&
+        where.position !== "after" &&
+        where.position !== "append")
+    ) {
+      throw new Error(
+        'drawioCreate: `where.position` must be one of "before", "after", "append"',
+      );
+    }
+    if (where.position === "before" || where.position === "after") {
+      const hasId =
+        typeof where.anchorNodeId === "string" && where.anchorNodeId.length > 0;
+      const hasText =
+        typeof where.anchorText === "string" && where.anchorText.length > 0;
+      if (hasId === hasText) {
+        throw new Error(
+          `drawioCreate: position "${where.position}" requires exactly one of anchorNodeId or anchorText`,
+        );
+      }
+    }
+
+    // Optional server-side ELK auto-layout: the model declares structure with
+    // rough coords, ELK computes the pixels (best-effort — returns the input
+    // unchanged on any layout failure).
+    const laidOutXml = layout === "elk" ? await applyElkLayout(xml) : xml;
+    // Pre-write pipeline (throws a structured DrawioLintError on any violation).
+    const prepared = prepareModel(laidOutXml);
+    const inner = renderDiagramShapes(prepared.cells, prepared.bbox);
+    const diagramTitle = title || "Page-1";
+    const svg = buildDrawioSvg(prepared.modelXml, inner, prepared.bbox, diagramTitle);
+
+    const att = await this.uploadAttachmentBuffer(
+      pageId,
+      Buffer.from(svg, "utf-8"),
+      "diagram.drawio.svg",
+      "image/svg+xml",
+    );
+
+    // NOTE: no `id` attribute is set here. The vendored `drawio` node schema
+    // (diagramAttributes) declares no `id`, so any block id would be silently
+    // dropped by PMNode.fromJSON on save and the returned handle would fail to
+    // resolve. The addressable handle is the node's "#<index>" (like image/table
+    // nodes), computed after the insert below.
+    const drawioNode: any = {
+      type: "drawio",
+      attrs: {
+        src: `/api/files/${att.id}/${att.fileName}`,
+        attachmentId: att.id,
+        width: prepared.bbox.width,
+        height: prepared.bbox.height,
+        align: "center",
+      },
+    };
+    if (title) drawioNode.attrs.title = title;
+    // Reuse the existing URL trust boundary (rejects unsafe src schemes).
+    this.validateDocUrls(drawioNode);
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    const pageUuid = await this.resolvePageId(pageId);
+
+    let inserted = false;
+    let insertedIndex = -1;
+    const mutation = await this.mutatePage(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        inserted = false;
+        insertedIndex = -1;
+        const { doc: nd, inserted: ins } = insertNodeRelative(
+          liveDoc,
+          drawioNode,
+          where,
+        );
+        inserted = ins;
+        if (!inserted) return null; // anchor not found -> skip the write
+        // Locate the freshly-inserted node to derive its "#<index>" handle. The
+        // just-uploaded attachmentId is unique, so it identifies our node.
+        if (Array.isArray(nd.content)) {
+          insertedIndex = nd.content.findIndex(
+            (b: any) =>
+              b &&
+              b.type === "drawio" &&
+              b.attrs &&
+              b.attrs.attachmentId === att.id,
+          );
+        }
+        return nd;
+      },
+    );
+
+    if (!inserted) {
+      const anchorDesc = where.anchorNodeId
+        ? `anchorNodeId "${where.anchorNodeId}"`
+        : `anchorText "${where.anchorText}"`;
+      throw new Error(
+        `drawioCreate: anchor not found (${anchorDesc}) on page ${pageId}. The diagram attachment ${att.id} is now an unreferenced orphan.`,
+      );
+    }
+
+    if (insertedIndex < 0) {
+      // The node was inserted nested (e.g. inside a callout/table cell via an
+      // anchor), where "#<index>" — which addresses only top-level blocks —
+      // cannot reference it. drawio nodes carry no persisted id, so there is no
+      // stable handle for a nested diagram.
+      throw new Error(
+        `drawioCreate: the diagram was inserted on page ${pageId} but not as a ` +
+          `top-level block, so it has no addressable "#<index>" handle. Anchor ` +
+          `on a top-level block (or append) so the diagram can be re-read.`,
+      );
+    }
+
+    // The returned handle is POSITIONAL ("#<index>"): valid for the immediate
+    // create -> get/update flow, but re-resolve via getOutline if the document
+    // structure changes (blocks added/removed before it shift the index).
+    const nodeId = `#${insertedIndex}`;
+
+    return {
+      success: true,
+      nodeId,
+      attachmentId: att.id,
+      warnings: prepared.warnings,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Full-replacement update of a drawio diagram. `baseHash` is MANDATORY: it is
+   * compared against the hash of the diagram's CURRENT XML (from drawioGet);
+   * any mismatch means a human or another agent edited the diagram after the
+   * read, so the write is refused with a conflict error. On success the new
+   * `.drawio.svg` is uploaded as a FRESH attachment (in-place byte overwrite is
+   * avoided — some Docmost versions corrupt an attachment on overwrite, exactly
+   * as replaceImage documents) and the node is repointed with new dimensions.
+   */
+  async drawioUpdate(
+    pageId: string,
+    node: string,
+    xml: string,
+    baseHash: string,
+    layout?: "elk",
+  ): Promise<{
+    success: boolean;
+    nodeId: string;
+    attachmentId: string;
+    warnings: string[];
+    verify?: any;
+  }> {
+    await this.ensureAuthenticated();
+    if (typeof baseHash !== "string" || baseHash.length === 0) {
+      throw new Error(
+        "drawioUpdate: baseHash is mandatory — read the diagram with drawioGet first and pass back its meta.hash",
+      );
+    }
+
+    // Resolve the node and read the CURRENT diagram to enforce the optimistic
+    // lock before doing any write or upload.
+    const { node: drawio, ref } = await this.resolveDrawioNode(pageId, node);
+    const oldAttrs = drawio.attrs || {};
+    const oldSrc = oldAttrs.src;
+    // The returned handle is the caller-supplied reference. drawio nodes carry
+    // no persisted id, so `ref` (an "#<index>" or a rare legacy attrs.id) is the
+    // honest identifier to hand back.
+    const nodeId = oldAttrs.id ?? ref;
+    if (!oldSrc) {
+      throw new Error(
+        `drawioUpdate: node "${node}" on page ${pageId} has no src to compare against`,
+      );
+    }
+    const currentSvg = await this.fetchAttachmentText(oldSrc);
+    const currentHash = mxHash(decodeDrawioSvg(currentSvg));
+    if (currentHash !== baseHash) {
+      throw new Error(
+        `drawioUpdate: conflict — the diagram changed since it was read ` +
+          `(baseHash ${baseHash} != current ${currentHash}). Re-read it with drawioGet and retry.`,
+      );
+    }
+
+    // Optional server-side ELK auto-layout (best-effort; see drawioCreate).
+    const laidOutXml = layout === "elk" ? await applyElkLayout(xml) : xml;
+    // Pipeline for the new content (throws a structured DrawioLintError).
+    const prepared = prepareModel(laidOutXml);
+    const inner = renderDiagramShapes(prepared.cells, prepared.bbox);
+    const diagramTitle = oldAttrs.title || "Page-1";
+    const svg = buildDrawioSvg(prepared.modelXml, inner, prepared.bbox, diagramTitle);
+
+    const att = await this.uploadAttachmentBuffer(
+      pageId,
+      Buffer.from(svg, "utf-8"),
+      "diagram.drawio.svg",
+      "image/svg+xml",
+    );
+    const newSrc = `/api/files/${att.id}/${att.fileName}`;
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    const pageUuid = await this.resolvePageId(pageId);
+
+    let repointed = 0;
+    const repoint = (n: any) => {
+      n.attrs = {
+        ...n.attrs,
+        src: newSrc,
+        attachmentId: att.id,
+        width: prepared.bbox.width,
+        height: prepared.bbox.height,
+      };
+      repointed++;
+    };
+
+    const mutation = await this.mutatePage(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        repointed = 0;
+        const doc =
+          liveDoc && liveDoc.type === "doc"
+            ? liveDoc
+            : { type: "doc", content: [] };
+        if (!Array.isArray(doc.content)) doc.content = [];
+        // Repoint ONLY the resolved node — never every node that happens to
+        // share this attachmentId (a copied diagram is two nodes with one
+        // attachmentId; keying on it would clobber both). Re-resolve the same
+        // handle against the live doc and walk to its exact position.
+        const hit = getNodeByRef(doc, ref);
+        if (!hit || hit.type !== "drawio") return null; // vanished/changed -> skip
+        let target: any = doc;
+        for (const idx of hit.path) {
+          if (!target || !Array.isArray(target.content)) {
+            target = null;
+            break;
+          }
+          target = target.content[idx];
+        }
+        if (!target || target.type !== "drawio") return null;
+        repoint(target);
+        if (repointed === 0) return null; // node vanished concurrently -> skip
+        return doc;
+      },
+    );
+
+    if (repointed === 0) {
+      return {
+        success: true,
+        nodeId,
+        attachmentId: att.id,
+        warnings: [
+          ...prepared.warnings,
+          "target drawio node was removed concurrently; uploaded attachment is unreferenced",
+        ],
+        verify: mutation.verify,
+      };
+    }
+
+    return {
+      success: true,
+      nodeId,
+      attachmentId: att.id,
+      warnings: prepared.warnings,
+      verify: mutation.verify,
+    };
+  }
+
+  // --- draw.io high-level semantic tools (issue #425) ---
+
+  /**
+   * ID-based targeted edits of an existing drawio diagram (add / update / delete
+   * cells) instead of resending the whole XML. Reads the CURRENT diagram, checks
+   * the optimistic lock (`baseHash` is MANDATORY, exactly as drawioUpdate), applies
+   * the operations to the parsed model (a `delete` CASCADES to container children
+   * and to every edge whose source/target is deleted), then runs the SAME #423
+   * pipeline as drawioUpdate (lint + quality warnings -> preview -> attachment ->
+   * repoint the node). Ids are stable so diffs stay meaningful across edits.
+   */
+
+  // --- draw.io high-level semantic tools (issue #425) ---
+
+  /**
+   * ID-based targeted edits of an existing drawio diagram (add / update / delete
+   * cells) instead of resending the whole XML. Reads the CURRENT diagram, checks
+   * the optimistic lock (`baseHash` is MANDATORY, exactly as drawioUpdate), applies
+   * the operations to the parsed model (a `delete` CASCADES to container children
+   * and to every edge whose source/target is deleted), then runs the SAME #423
+   * pipeline as drawioUpdate (lint + quality warnings -> preview -> attachment ->
+   * repoint the node). Ids are stable so diffs stay meaningful across edits.
+   */
+  async drawioEditCells(
+    pageId: string,
+    node: string,
+    operations: CellOp[],
+    baseHash: string,
+  ): Promise<{
+    success: boolean;
+    nodeId: string;
+    attachmentId: string;
+    warnings: string[];
+    verify?: any;
+  }> {
+    await this.ensureAuthenticated();
+    if (typeof baseHash !== "string" || baseHash.length === 0) {
+      throw new Error(
+        "drawioEditCells: baseHash is mandatory — read the diagram with drawioGet first and pass back its meta.hash",
+      );
+    }
+    if (!Array.isArray(operations) || operations.length === 0) {
+      throw new Error(
+        "drawioEditCells: operations must be a non-empty array of { op, ... }",
+      );
+    }
+
+    const { node: drawio, ref } = await this.resolveDrawioNode(pageId, node);
+    const oldAttrs = drawio.attrs || {};
+    const oldSrc = oldAttrs.src;
+    const nodeId = oldAttrs.id ?? ref;
+    if (!oldSrc) {
+      throw new Error(
+        `drawioEditCells: node "${node}" on page ${pageId} has no src to edit`,
+      );
+    }
+    const currentSvg = await this.fetchAttachmentText(oldSrc);
+    const currentModel = decodeDrawioSvg(currentSvg);
+    const currentHash = mxHash(currentModel);
+    if (currentHash !== baseHash) {
+      throw new Error(
+        `drawioEditCells: conflict — the diagram changed since it was read ` +
+          `(baseHash ${baseHash} != current ${currentHash}). Re-read it with drawioGet and retry.`,
+      );
+    }
+
+    // Apply the operations to the parsed model, then run the standard pipeline.
+    const editedModel = applyCellOps(currentModel, operations);
+    const prepared = prepareModel(editedModel);
+    const inner = renderDiagramShapes(prepared.cells, prepared.bbox);
+    const diagramTitle = oldAttrs.title || "Page-1";
+    const svg = buildDrawioSvg(prepared.modelXml, inner, prepared.bbox, diagramTitle);
+
+    const att = await this.uploadAttachmentBuffer(
+      pageId,
+      Buffer.from(svg, "utf-8"),
+      "diagram.drawio.svg",
+      "image/svg+xml",
+    );
+    const newSrc = `/api/files/${att.id}/${att.fileName}`;
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    const pageUuid = await this.resolvePageId(pageId);
+
+    let repointed = 0;
+    const mutation = await this.mutatePage(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        repointed = 0;
+        const doc =
+          liveDoc && liveDoc.type === "doc" ? liveDoc : { type: "doc", content: [] };
+        if (!Array.isArray(doc.content)) doc.content = [];
+        const hit = getNodeByRef(doc, ref);
+        if (!hit || hit.type !== "drawio") return null;
+        let target: any = doc;
+        for (const idx of hit.path) {
+          if (!target || !Array.isArray(target.content)) {
+            target = null;
+            break;
+          }
+          target = target.content[idx];
+        }
+        if (!target || target.type !== "drawio") return null;
+        target.attrs = {
+          ...target.attrs,
+          src: newSrc,
+          attachmentId: att.id,
+          width: prepared.bbox.width,
+          height: prepared.bbox.height,
+        };
+        repointed++;
+        return doc;
+      },
+    );
+
+    if (repointed === 0) {
+      return {
+        success: true,
+        nodeId,
+        attachmentId: att.id,
+        warnings: [
+          ...prepared.warnings,
+          "target drawio node was removed concurrently; uploaded attachment is unreferenced",
+        ],
+        verify: mutation.verify,
+      };
+    }
+    return {
+      success: true,
+      nodeId,
+      attachmentId: att.id,
+      warnings: prepared.warnings,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * The main high-level tool: build a diagram from a SEMANTIC graph (nodes with
+   * a `kind`/`icon`, groups, edges) — the model never supplies coordinates or
+   * style strings. The server resolves icons via the shape catalog (#424),
+   * assigns palette colors from the preset, runs ELK layered layout (honouring
+   * `direction` and the `layer`/`sameLayerAs`/`pinned` hints and compound groups),
+   * and assembles linter-clean XML, then inserts it through the SAME create
+   * pipeline as drawioCreate. `layout:"incremental"` is only meaningful when a
+   * target `node` is given (it preserves that diagram's existing coordinates and
+   * places only new cells); on a fresh insert it behaves like "full".
+   */
+  async drawioFromGraph(
+    pageId: string,
+    where: {
+      position: "before" | "after" | "append";
+      anchorNodeId?: string;
+      anchorText?: string;
+    },
+    graph: Graph,
+    direction?: "LR" | "RL" | "TB" | "BT",
+    preset?: string,
+    layout?: GraphLayoutMode,
+    node?: string,
+  ): Promise<{
+    success: boolean;
+    nodeId: string;
+    attachmentId: string;
+    warnings: string[];
+    iconsResolved: number;
+    iconsMissing: string[];
+    verify?: any;
+  }> {
+    await this.ensureAuthenticated();
+    // Direction/preset supplied as separate params override the graph fields so
+    // both the flat tool schema and an inline graph can set them.
+    const merged: Graph = {
+      ...graph,
+      direction: direction ?? graph.direction,
+      preset: preset ?? graph.preset,
+    };
+    const mode: GraphLayoutMode = layout ?? "full";
+
+    // Incremental into an EXISTING node: read its coords so ELK preserves them,
+    // and keep the full existing model so incremental MERGES (never drops) any
+    // cell the new graph doesn't re-list.
+    let existingCoords: Map<string, { x: number; y: number }> | undefined;
+    let existingModelXml: string | undefined;
+    let editExisting = false;
+    let baseHash: string | undefined;
+    if (node && (mode === "incremental" || mode === "none")) {
+      const { node: drawio } = await this.resolveDrawioNode(pageId, node);
+      const src = (drawio.attrs || {}).src;
+      if (src) {
+        const svg = await this.fetchAttachmentText(src);
+        const model = decodeDrawioSvg(svg);
+        baseHash = mxHash(model);
+        existingModelXml = model;
+        existingCoords = new Map();
+        for (const c of parseDrawioCells(model)) {
+          if (c.vertex && c.geometry.x != null && c.geometry.y != null) {
+            existingCoords.set(c.id, { x: c.geometry.x, y: c.geometry.y });
+          }
+        }
+        editExisting = true;
+      }
+    }
+
+    const built = await buildFromGraph(
+      merged,
+      mode,
+      existingCoords,
+      existingModelXml,
+    );
+
+    if (editExisting && node && baseHash) {
+      // Re-target the existing diagram: replace it with the assembled model.
+      const res = await this.drawioUpdate(pageId, node, built.modelXml, baseHash);
+      return {
+        ...res,
+        iconsResolved: built.iconsResolved,
+        iconsMissing: built.iconsMissing,
+      };
+    }
+
+    const res = await this.drawioCreate(pageId, where, built.modelXml);
+    return {
+      ...res,
+      iconsResolved: built.iconsResolved,
+      iconsMissing: built.iconsMissing,
+    };
+  }
+
+  /**
+   * Convert a Mermaid `flowchart` to a redactable draw.io diagram via a PURE
+   * parser (no Electron / draw.io CLI): mermaid text -> graph-JSON -> the
+   * drawioFromGraph pipeline. Only `flowchart`/`graph` is supported (the most
+   * common wiki case); other diagram types throw a clear error so the model can
+   * fall back to drawioFromGraph.
+   */
+  async drawioFromMermaid(
+    pageId: string,
+    where: {
+      position: "before" | "after" | "append";
+      anchorNodeId?: string;
+      anchorText?: string;
+    },
+    mermaid: string,
+    preset?: string,
+  ): Promise<{
+    success: boolean;
+    nodeId: string;
+    attachmentId: string;
+    warnings: string[];
+    iconsResolved: number;
+    iconsMissing: string[];
+    verify?: any;
+  }> {
+    await this.ensureAuthenticated();
+    const graph = mermaidToGraph(mermaid);
+    if (preset) graph.preset = preset;
+    return this.drawioFromGraph(pageId, where, graph, graph.direction, graph.preset);
+  }
+
+  // --- Page history / diff / transform ---
+
+  /**
+   * List the saved versions (history snapshots) of a page, newest first.
+   * Docmost auto-snapshots on every save. Returns one cursor-paginated page of
+   * results: `{ items, nextCursor }`. The history record's id field is `id`.
+   */
+  }
+  return DrawioMixin;
+}

packages/mcp/src/client/errors.ts

@@ -0,0 +1,165 @@
+// Central REST error diagnostics (issues #437 + #450). SINGLE place that maps an
+// axios error to the model-facing message. Extracted verbatim from client.ts;
+// the constructor's response interceptor (see client/context.ts) routes every
+// REST call through formatDocmostAxiosError so the whole surface is uniform.
+import axios from "axios";
+
+// --- Issue #437: central error diagnostics -------------------------------
+// The agent only ever sees the thrown exception's `error.message`, so a failed
+// tool must return an ACTIONABLE message (method, path, status, and the
+// server's own validation text) instead of the opaque "Request failed with
+// status code 400". These helpers + the response interceptor in the
+// constructor are the single authoritative place that text is composed.
+
+// Overall cap on the composed diagnostic message so the model context stays
+// compact and a (whitelisted) server string can never blow up the text.
+const ERROR_MESSAGE_CAP = 300;
+// Only attempt to JSON.parse an arraybuffer body under this size: a larger
+// binary body is never a JSON error envelope, so parsing it just wastes memory
+// (fetchInternalFile uses responseType:"arraybuffer", so a failed file fetch
+// carries the JSON error envelope as raw bytes here).
+const ERROR_BUFFER_PARSE_CAP = 4096;
+
+// Canonical 36-char UUID (8-4-4-4-12 hex). Deliberately version/variant-
+// AGNOSTIC: the ids are UUIDv7 (e.g. 019f499a-9f8c-7d68-...), so only the
+// canonical shape/length is enforced, not the version/variant nibble.
+const FULL_UUID_RE =
+  /^[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}$/i;
+
+/**
+ * Throw an actionable error BEFORE any network call when `value` is not a full
+ * canonical UUID. Absorbs #436: a truncated/short comment id used to reach the
+ * server and bounce back as an opaque 400/404 the agent could not self-correct;
+ * failing fast here names the exact fix.
+ */
+export function assertFullUuid(
+  tool: string,
+  param: string,
+  value: string,
+): void {
+  if (typeof value !== "string" || !FULL_UUID_RE.test(value)) {
+    throw new Error(
+      `${tool}: '${param}' must be the FULL comment UUID (36 chars, e.g. ` +
+        `019f499a-9f8c-7d68-b7be-ce100d7c6c56), got '${value}'. Copy the id ` +
+        `verbatim from listComments / createComment output.`,
+    );
+  }
+}
+
+// Keep ONLY the pathname of a request (no host, no query string, no fragment)
+// so the message never leaks a host or query params. Resolves a relative
+// config.url against config.baseURL, then discards everything but the path.
+function requestPath(config: any): string {
+  const rawUrl = typeof config?.url === "string" ? config.url : "";
+  const base =
+    typeof config?.baseURL === "string" ? config.baseURL : undefined;
+  try {
+    // A dummy base makes an absolute config.url parse too; its host is dropped.
+    return new URL(rawUrl, base ?? "http://localhost").pathname;
+  } catch {
+    // Malformed url: still strip any query/fragment manually.
+    return rawUrl.split(/[?#]/)[0] || rawUrl;
+  }
+}
+
+/**
+ * Compose the server-facing message from `error.response.data`, using ONLY the
+ * whitelisted `message`/`error` fields or the HTTP statusText. SECURITY: the
+ * raw response body, headers (Authorization!) and config are NEVER read here —
+ * a string/HTML body (e.g. a proxy's 502 page) is deliberately dropped in
+ * favour of the statusText.
+ */
+function extractServerMessage(data: any, statusText: string): string {
+  // class-validator envelope: { message: string | string[], error?: string }.
+  if (
+    data &&
+    typeof data === "object" &&
+    !Buffer.isBuffer(data) &&
+    !(data instanceof ArrayBuffer)
+  ) {
+    const msg = (data as any).message;
+    if (Array.isArray(msg)) {
+      const joined = msg.filter((m) => typeof m === "string").join("; ");
+      if (joined) return joined;
+    } else if (typeof msg === "string" && msg) {
+      return msg;
+    }
+    const err = (data as any).error;
+    if (typeof err === "string" && err) return err;
+    return statusText;
+  }
+
+  // Buffer / ArrayBuffer body: attempt a size-capped, guarded JSON.parse so a
+  // failed arraybuffer fetch still surfaces the server's validation text.
+  if (Buffer.isBuffer(data) || data instanceof ArrayBuffer) {
+    const buf = Buffer.isBuffer(data) ? data : Buffer.from(data);
+    if (buf.length > 0 && buf.length <= ERROR_BUFFER_PARSE_CAP) {
+      try {
+        return extractServerMessage(JSON.parse(buf.toString("utf8")), statusText);
+      } catch {
+        return statusText;
+      }
+    }
+    return statusText;
+  }
+
+  // A raw string / HTML body is never surfaced (may echo server internals).
+  return statusText;
+}
+
+/**
+ * Reformat an AxiosError's `.message` IN PLACE into an actionable diagnostic:
+ *   `<METHOD> <path> failed (<status> <statusText>): <serverMessage>`
+ * or, when the request never got a response:
+ *   `<METHOD> <path> failed: <code> (no response from server)`.
+ *
+ * Mutates the SAME error object (never a custom subclass) so the live
+ * axios.isAxiosError / error.response?.status / config._retry checks around the
+ * client keep working, and sets `_docmostFormatted` as a double-processing
+ * guard. A no-op on a non-axios or already-formatted error.
+ */
+export function formatDocmostAxiosError(error: any): void {
+  if (!error || error._docmostFormatted) return;
+  if (!axios.isAxiosError(error)) return;
+
+  const config: any = error.config ?? {};
+  const method =
+    typeof config.method === "string" ? config.method.toUpperCase() : "";
+  const methodPath = `${method} ${requestPath(config)}`.trim();
+  const response = error.response;
+
+  let message: string;
+  if (response) {
+    const statusText =
+      typeof response.statusText === "string" ? response.statusText : "";
+    const serverMessage = extractServerMessage(response.data, statusText);
+    message = `${methodPath} failed (${response.status} ${statusText}): ${serverMessage}`;
+    // Full body only to stderr under DEBUG (parity with downloadImage).
+    if (process.env.DEBUG) {
+      console.error(
+        "Docmost request failed; response body:",
+        JSON.stringify(response.data),
+      );
+    }
+  } else {
+    // No response at all (ECONNREFUSED / ETIMEDOUT / ECONNRESET / DNS / timeout).
+    // Use ONLY error.code, never the raw error.message: axios network messages
+    // embed host:port ("connect ECONNREFUSED 127.0.0.1:3000", "getaddrinfo
+    // ENOTFOUND host") and #437's invariant is that the host never reaches the
+    // model-visible message. code is set for essentially every real no-response
+    // error (ECONNREFUSED/ETIMEDOUT/ECONNRESET/ENOTFOUND/ECONNABORTED); the full
+    // native message still goes to stderr under DEBUG.
+    const reason = error.code ?? "network error";
+    message = `${methodPath} failed: ${reason} (no response from server)`;
+    if (process.env.DEBUG) {
+      console.error("Docmost request failed; no response:", error.message);
+    }
+  }
+
+  if (message.length > ERROR_MESSAGE_CAP) {
+    message = message.slice(0, ERROR_MESSAGE_CAP - 1) + "…";
+  }
+
+  error.message = message;
+  (error as any)._docmostFormatted = true;
+}

packages/mcp/src/client/getpage-cache.ts

@@ -0,0 +1,147 @@
+// Content-addressed LRU cache for the PM->Markdown conversion in getPage
+// (issue #479). getPage is the dominant agent op (812 calls / 2h, p95 840ms);
+// the bulk of its cost is convertProseMirrorToMarkdown — a full ProseMirror-tree
+// walk over the page content (hundreds of KB of JSON on large pages) run on
+// EVERY read. Since agents re-read far more than they write (812 reads vs 28
+// writes in the sample), most conversions re-produce the SAME markdown from
+// UNCHANGED content. This cache skips the recomputation on a hit.
+//
+// KEY = (pageId, updatedAt, optionsHash):
+//   - pageId: the page's CANONICAL UUID (resultData.id), not the agent-supplied
+//     slugId — so a slugId read and a UUID read of the same page share one entry.
+//   - updatedAt: comes from the SAME /pages/info response as `content`, so the
+//     two are mutually consistent; a changed page yields a new updatedAt -> a new
+//     key -> automatic, precise invalidation (no stale markdown is ever served).
+//   - optionsHash: a stable hash of the conversion options. getPage passes
+//     `{dropResolvedCommentAnchors:true}` while exportPageMarkdown passes the
+//     defaults (#328) — DIFFERENT output for the same content, so the options
+//     MUST be part of the key or a hit would serve the wrong variant.
+//
+// BOUNDS: evict the LEAST-recently-used entry when EITHER the entry count OR the
+// total stored bytes would exceed its cap. Large pages are hundreds of KB, so a
+// byte cap (not just a count cap) is what actually bounds memory. A Map iterates
+// in insertion order, so the first key is the LRU entry; a hit re-inserts its key
+// to move it to the most-recently-used end.
+//
+// This module is dependency-neutral (no axios/client/prom-client): a plain class
+// the shared client context owns one instance of, so the cache persists across
+// getPage calls on a single DocmostClient instance (built per user / per chat).
+
+/** A stable, order-insensitive hash of the conversion options object. */
+export function hashConvertOptions(options: unknown): string {
+  // JSON.stringify with SORTED keys makes the hash independent of key order, so
+  // {a:1,b:2} and {b:2,a:1} collapse to one entry. undefined/null options -> a
+  // fixed empty-object key, matching a caller that passes no options at all.
+  if (options === undefined || options === null) return "{}";
+  return stableStringify(options);
+}
+
+function stableStringify(value: any): string {
+  if (value === null || typeof value !== "object") return JSON.stringify(value);
+  if (Array.isArray(value)) return `[${value.map(stableStringify).join(",")}]`;
+  const keys = Object.keys(value).sort();
+  return `{${keys
+    .map((k) => `${JSON.stringify(k)}:${stableStringify(value[k])}`)
+    .join(",")}}`;
+}
+
+interface CacheEntry {
+  markdown: string;
+  bytes: number;
+}
+
+export interface GetPageCacheOptions {
+  /** Max number of entries before LRU eviction. Default 50. */
+  maxEntries?: number;
+  /** Max total stored bytes before LRU eviction. Default 10 MB. */
+  maxBytes?: number;
+}
+
+export class GetPageConversionCache {
+  private readonly maxEntries: number;
+  private readonly maxBytes: number;
+  // Insertion-ordered: the FIRST key is the least-recently-used entry.
+  private readonly map = new Map<string, CacheEntry>();
+  private totalBytes = 0;
+
+  constructor(opts: GetPageCacheOptions = {}) {
+    // A non-positive/NaN cap is treated as "use the default", never as an
+    // unbounded (or always-empty) cache — a silently unbounded cache would leak
+    // memory, and an always-empty one would defeat the whole optimization.
+    this.maxEntries =
+      Number.isFinite(opts.maxEntries) && (opts.maxEntries as number) > 0
+        ? Math.floor(opts.maxEntries as number)
+        : 50;
+    this.maxBytes =
+      Number.isFinite(opts.maxBytes) && (opts.maxBytes as number) > 0
+        ? Math.floor(opts.maxBytes as number)
+        : 10 * 1024 * 1024;
+  }
+
+  /** Compose the content-addressed key from its three parts. */
+  static key(pageId: string, updatedAt: string, optionsHash: string): string {
+    // A space separates the parts so no combination of values can collide by
+    // concatenation: a canonical UUID and an ISO updatedAt never contain a
+    // space, so the boundaries between the three parts are unambiguous.
+    return `${pageId} ${updatedAt} ${optionsHash}`;
+  }
+
+  /**
+   * Return the cached markdown for `key`, or undefined on a miss. A hit moves
+   * the entry to the most-recently-used end (delete + re-set) so the LRU order
+   * reflects real access, not just insertion.
+   */
+  get(key: string): string | undefined {
+    const entry = this.map.get(key);
+    if (entry === undefined) return undefined;
+    this.map.delete(key);
+    this.map.set(key, entry);
+    return entry.markdown;
+  }
+
+  /**
+   * Store `markdown` under `key`, then evict LRU entries until BOTH caps hold.
+   * Re-storing an existing key refreshes its value and recency (its old bytes
+   * are subtracted first, so totalBytes stays exact).
+   */
+  set(key: string, markdown: string): void {
+    // Byte size of the stored string (UTF-8). A single entry larger than the
+    // whole byte cap is still stored (so getPage always gets a hit next time),
+    // then the eviction loop below simply cannot shrink below it — accepted:
+    // one oversized page is bounded by the page itself, not a cache leak.
+    const bytes = Buffer.byteLength(markdown, "utf8");
+    const existing = this.map.get(key);
+    if (existing !== undefined) {
+      this.totalBytes -= existing.bytes;
+      this.map.delete(key);
+    }
+    this.map.set(key, { markdown, bytes });
+    this.totalBytes += bytes;
+    this.evict();
+  }
+
+  /** Evict the LRU entry until both the count and byte caps are satisfied. */
+  private evict(): void {
+    while (
+      this.map.size > this.maxEntries ||
+      (this.totalBytes > this.maxBytes && this.map.size > 1)
+    ) {
+      // The first key in insertion order is the least-recently-used.
+      const oldest = this.map.keys().next().value as string | undefined;
+      if (oldest === undefined) break;
+      const entry = this.map.get(oldest);
+      this.map.delete(oldest);
+      if (entry) this.totalBytes -= entry.bytes;
+    }
+  }
+
+  /** Current entry count (test/introspection). */
+  get size(): number {
+    return this.map.size;
+  }
+
+  /** Current total stored bytes (test/introspection). */
+  get bytes(): number {
+    return this.totalBytes;
+  }
+}

packages/mcp/src/client/media.ts

@@ -0,0 +1,730 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import FormData from "form-data";
+import axios, { AxiosInstance } from "axios";
+import { basename, extname } from "path";
+import {
+  updatePageContentRealtime,
+  replacePageContent,
+  markdownToProseMirror,
+  markdownToProseMirrorCanonical,
+  mutatePageContent,
+  assertYjsEncodable,
+  MutationResult,
+} from "../lib/collaboration.js";
+import { withPageLock, isUuid } from "../lib/page-lock.js";
+import { diffDocs, summarizeChange } from "../lib/diff.js";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+
+// Supported image types, kept as two lookup tables so both a local file
+// extension and a remote Content-Type can be mapped to the same canonical set.
+const EXT_TO_MIME: Record<string, string> = {
+  ".png": "image/png",
+  ".jpg": "image/jpeg",
+  ".jpeg": "image/jpeg",
+  ".gif": "image/gif",
+  ".webp": "image/webp",
+  ".svg": "image/svg+xml",
+};
+const MIME_TO_EXT: Record<string, string> = {
+  "image/png": ".png",
+  "image/jpeg": ".jpg",
+  "image/gif": ".gif",
+  "image/webp": ".webp",
+  "image/svg+xml": ".svg",
+};
+
+// Public method surface of MediaMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements IMediaMixin` fails to compile on drift.
+export interface IMediaMixin {
+  uploadImage(pageId: string, url: string): any;
+  insertImage(pageId: string, url: string, opts?: { align?: "left" | "center" | "right"; alt?: string; replaceText?: string; afterText?: string; }): any;
+  replaceImage(pageId: string, oldAttachmentId: string, url: string, opts?: { align?: "left" | "center" | "right"; alt?: string }): any;
+}
+
+export function MediaMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & IMediaMixin> & TBase {
+  abstract class MediaMixin extends Base implements IMediaMixin {
+  // --- Image upload / embedding ---
+
+  /** Map a Content-Type string to a supported MIME type, or null if unsupported. */
+  protected supportedImageMime(ct: string): string | null {
+    return MIME_TO_EXT[ct] ? ct : null;
+  }
+
+  /**
+   * Download a remote image from a caller-supplied URL and resolve its bytes,
+   * MIME and a filename.
+   *
+   * SSRF / RESOURCE TRUST BOUNDARY: the URL comes from the MCP caller and is
+   * fetched BY THE SERVER, so it must be guarded before and after the request.
+   * The guards mirror the local-file trust boundary in uploadImage:
+   *   - scheme allowlist (http/https only) — rejects file:, data:, ftp:, etc.,
+   *     so the caller cannot use this path to read local files or other schemes;
+   *   - a size cap enforced both via axios maxContentLength/maxBodyLength AND a
+   *     post-download buffer.length re-check (defends against a missing/lying
+   *     Content-Length), so a huge response cannot exhaust memory;
+   *   - a 30s timeout. The timeout matters because replaceImage holds the
+   *     per-page lock across this upload, so a hung download would wedge the
+   *     lock for that page.
+   * We deliberately do NOT block private IP ranges: the MCP caller is already
+   * trusted to read arbitrary host files via the filePath path, so the marginal
+   * trust granted by fetching internal URLs is comparable, and blocking would
+   * break legitimate internal-image use.
+   */
+  protected async fetchRemoteImage(
+    url: string,
+    maxBytes: number,
+  ): Promise<{ buffer: Buffer; mime: string; fileName: string }> {
+    // Scheme allowlist first — cheapest guard, and rejects non-http(s) schemes
+    // (file:, data:, ftp:, ...) before any network request is made.
+    let parsed: URL;
+    try {
+      parsed = new URL(url);
+    } catch (e: any) {
+      throw new Error(`Invalid image URL "${url}": ${e.message}`);
+    }
+    if (parsed.protocol !== "http:" && parsed.protocol !== "https:") {
+      throw new Error(
+        `unsupported image URL scheme "${parsed.protocol}"; only http and https are allowed`,
+      );
+    }
+
+    let response;
+    try {
+      response = await axios.get(url, {
+        responseType: "arraybuffer",
+        timeout: 30000,
+        maxContentLength: maxBytes,
+        maxBodyLength: maxBytes,
+        headers: { Accept: "image/*" },
+      });
+    } catch (error) {
+      // Keep the thrown message free of the raw response body (it may echo
+      // server internals); surface only status/statusText. The full body is
+      // logged under DEBUG for diagnostics.
+      if (axios.isAxiosError(error)) {
+        if (process.env.DEBUG) {
+          console.error(
+            "Image download failed; response body:",
+            JSON.stringify(error.response?.data),
+          );
+        }
+        throw new Error(
+          `Image download failed for "${url}": ${error.response?.status ?? ""} ${error.response?.statusText ?? error.message}`.trim(),
+        );
+      }
+      throw error;
+    }
+
+    // axios returns an ArrayBuffer for responseType: "arraybuffer".
+    const buffer = Buffer.from(response.data);
+    // Re-check the size: maxContentLength relies on Content-Length, which may be
+    // absent or lie, so guard against the actual byte count too.
+    if (buffer.length === 0) {
+      throw new Error(`Empty image response from "${url}"`);
+    }
+    if (buffer.length > maxBytes) {
+      throw new Error(
+        `Image too large: ${buffer.length} bytes exceeds the ${maxBytes}-byte cap`,
+      );
+    }
+
+    // Resolve MIME: prefer the response Content-Type (strip any "; charset=..."
+    // parameter, lowercase, trim) mapped through the supported set; if the
+    // header is generic/missing/unsupported, fall back to the URL path
+    // extension via the existing extension->MIME logic.
+    const rawCt = response.headers?.["content-type"];
+    let mime: string | null = null;
+    if (typeof rawCt === "string" && rawCt.length > 0) {
+      const ct = rawCt.split(";")[0].trim().toLowerCase();
+      mime = this.supportedImageMime(ct);
+    }
+    if (!mime) {
+      // Fall back to the URL path extension. Use the pathname so the query
+      // string never contaminates the extension lookup.
+      const ext = extname(parsed.pathname).toLowerCase();
+      mime = EXT_TO_MIME[ext] ?? null;
+    }
+    if (!mime) {
+      throw new Error(
+        `cannot determine supported image type for "${url}"; supported: png, jpg, jpeg, gif, webp, svg`,
+      );
+    }
+
+    // Build a filename from the URL path basename (ignore the query string),
+    // defaulting to "image" when empty, and ensure it ends with the canonical
+    // extension for the resolved MIME (append it when missing/mismatched).
+    const canonicalExt = MIME_TO_EXT[mime];
+    let fileName = basename(parsed.pathname) || "image";
+    if (extname(fileName).toLowerCase() !== canonicalExt) {
+      fileName += canonicalExt;
+    }
+
+    return { buffer, mime, fileName };
+  }
+
+  /** Build a Docmost ProseMirror image node from an uploaded attachment. */
+  protected buildImageNode(
+    att: { id: string; fileName: string; fileSize?: number },
+    align?: "left" | "center" | "right",
+    alt?: string,
+  ): any {
+    // Clean file URL, matching Docmost's native behaviour. No cache-busting
+    // query: the server serves the bare URL correctly, and replacement creates
+    // a new attachment id (a new URL) which busts caches naturally.
+    const src = `/api/files/${att.id}/${att.fileName}`;
+    const node: any = {
+      type: "image",
+      attrs: {
+        src,
+        attachmentId: att.id,
+        // Default to null when the server omits fileSize so the attr is never
+        // undefined (undefined would be dropped on serialization / break the
+        // ProseMirror image schema which expects size present).
+        size: att.fileSize ?? null,
+        align: align || "center",
+        width: null,
+      },
+    };
+    if (alt) node.attrs.alt = alt;
+    return node;
+  }
+
+  /**
+   * Download a remote image from an http(s) URL and upload it as an attachment
+   * of a page, returning the attachment metadata plus a ready-to-insert
+   * ProseMirror image node. Local file paths are intentionally not supported:
+   * the MCP caller is a remote AI with no access to this server's filesystem.
+   */
+  async uploadImage(pageId: string, url: string) {
+    await this.ensureAuthenticated();
+
+    const MAX_IMAGE_BYTES = 20 * 1024 * 1024; // 20 MiB
+
+    // Fetch + validate the remote image (scheme allowlist, size cap, timeout).
+    // See fetchRemoteImage for the SSRF / resource trust boundary.
+    const fetched = await this.fetchRemoteImage(url, MAX_IMAGE_BYTES);
+    const fileBuffer = fetched.buffer;
+    const mime = fetched.mime;
+    const fileName = fetched.fileName;
+
+    // Build a FRESH FormData for every send attempt. A FormData body is a
+    // single-use stream that is CONSUMED on the first send, so it cannot be
+    // replayed by this.client's response interceptor (replaying a consumed
+    // stream fails with 'socket hang up'). Multipart re-auth is therefore done
+    // here with bare axios and an explicit one-shot 401/403 retry that rebuilds
+    // the body. Field order matters: text fields must precede the file part so
+    // the server reads them; the server always generates a fresh attachment id.
+    const buildForm = () => {
+      const form = new FormData();
+      form.append("pageId", pageId);
+      form.append("file", fileBuffer, {
+        filename: fileName,
+        contentType: mime,
+      });
+      return form;
+    };
+
+    // Local name distinct from the `url` parameter (the source image URL): this
+    // is the /files/upload endpoint we POST the multipart body to.
+    const uploadUrl = `${this.apiUrl}/files/upload`;
+    let response;
+    try {
+      // Call buildForm() ONCE per attempt and reuse the instance for both
+      // getHeaders() and the body so the Content-Type boundary matches the body.
+      const form = buildForm();
+      // Read the Authorization header from this.client's defaults (set by
+      // login(), only ever deleted — never set to null) instead of building
+      // `Bearer ${this.token}`: a concurrent JSON 401 can null this.token
+      // mid-flight, which would otherwise produce a literal "Bearer null".
+      // ensureAuthenticated() above guarantees login() ran, so the default
+      // header exists here. A 60s timeout keeps a hung upload from wedging the
+      // per-page lock (replaceImage holds withPageLock across this call).
+      response = await axios.post(uploadUrl, form, {
+        headers: {
+          ...form.getHeaders(),
+          Authorization: this.client.defaults.headers.common["Authorization"],
+        },
+        timeout: 60000,
+      });
+    } catch (error) {
+      // On an expired-token auth error, re-login and retry exactly once with a
+      // freshly-rebuilt FormData (the previous one was already consumed).
+      if (
+        axios.isAxiosError(error) &&
+        (error.response?.status === 401 || error.response?.status === 403)
+      ) {
+        await this.login();
+        const form2 = buildForm();
+        response = await axios.post(uploadUrl, form2, {
+          headers: {
+            ...form2.getHeaders(),
+            Authorization: this.client.defaults.headers.common["Authorization"],
+          },
+          timeout: 60000,
+        });
+      } else if (axios.isAxiosError(error)) {
+        // Keep the thrown message free of the raw response body (it may echo
+        // request data or server internals); surface only status/statusText.
+        // The full body is logged under DEBUG for diagnostics.
+        if (process.env.DEBUG) {
+          console.error(
+            "Image upload failed; response body:",
+            JSON.stringify(error.response?.data),
+          );
+        }
+        throw new Error(
+          `Image upload failed: ${error.response?.status} ${error.response?.statusText}`,
+        );
+      } else {
+        throw error;
+      }
+    }
+    // The attachment may arrive bare or wrapped in a { data } envelope.
+    const att = response.data?.data ?? response.data;
+    if (!att?.id || !att?.fileName) {
+      throw new Error(
+        "Unexpected /files/upload response: " + JSON.stringify(response.data),
+      );
+    }
+
+    // Some Docmost versions omit fileSize from the upload response. Fall back
+    // to the fetched byte length (the bytes we just uploaded) so callers never
+    // get an undefined size.
+    const resolvedSize = att.fileSize ?? fileBuffer.length;
+
+    return {
+      attachmentId: att.id,
+      fileName: att.fileName,
+      fileSize: resolvedSize,
+      src: `/api/files/${att.id}/${att.fileName}`,
+      imageNode: this.buildImageNode({ ...att, fileSize: resolvedSize }),
+    };
+  }
+
+  /**
+   * Upload an image from a web (http/https) URL and insert it into a page in
+   * one step.
+   * By default the image is appended at the end. With replaceText, the first
+   * top-level block whose text contains the string is replaced; with afterText,
+   * the image is inserted right after the first matching block. All other
+   * block ids are preserved (only one top-level block is added or swapped).
+   */
+  async insertImage(
+    pageId: string,
+    url: string,
+    opts: {
+      align?: "left" | "center" | "right";
+      alt?: string;
+      replaceText?: string;
+      afterText?: string;
+    } = {},
+  ) {
+    const up = await this.uploadImage(pageId, url);
+    // Reuse the node from uploadImage (clean /api/files/<id>/<file> src), then
+    // apply align/alt onto a shallow attrs copy.
+    const node: any = { ...up.imageNode, attrs: { ...up.imageNode.attrs } };
+    if (opts.align) node.attrs.align = opts.align;
+    if (opts.alt) node.attrs.alt = opts.alt;
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260). The
+    // uploadImage /files/upload call above keeps the agent-supplied id.
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Recursively collect the plain text of a top-level block.
+    const blockText = (n: any): string => {
+      let out = "";
+      if (n.type === "text") out += n.text || "";
+      for (const child of n.content || []) out += blockText(child);
+      return out;
+    };
+
+    // Insert into the LIVE synced document, not the debounced REST snapshot, so
+    // concurrent edits/comments/images are preserved and parallel insertImage
+    // calls (serialized by the per-page lock) each see the previous insertion.
+    let placement: "replaced" | "after" | "appended" | undefined;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        const doc =
+          liveDoc && liveDoc.type === "doc"
+            ? liveDoc
+            : { type: "doc", content: [] };
+        if (!Array.isArray(doc.content)) doc.content = [];
+
+        if (opts.replaceText) {
+          // Ambiguity guard (mirrors editPageText): count matching top-level
+          // blocks first, so a non-unique fragment cannot silently replace the
+          // wrong block (e.g. text that also appears inside a callout/table).
+          const matches = doc.content.filter((b: any) =>
+            blockText(b).includes(opts.replaceText!),
+          );
+          if (matches.length === 0) {
+            throw new Error(`replaceText not found: "${opts.replaceText}"`);
+          }
+          if (matches.length > 1) {
+            throw new Error(
+              `replaceText "${opts.replaceText}" matches ${matches.length} blocks; use a longer unique fragment`,
+            );
+          }
+          const idx = doc.content.findIndex((b: any) =>
+            blockText(b).includes(opts.replaceText!),
+          );
+          // Data-loss guard: replaceText swaps the WHOLE top-level block, so if
+          // the fragment only appears nested inside a container (table, callout,
+          // list, blockquote) the entire structure would be destroyed. Refuse
+          // when the matched block is a container rather than a leaf
+          // paragraph/heading and point the caller at a safer tool.
+          const CONTAINER_TYPES = new Set([
+            "table",
+            "callout",
+            "bulletList",
+            "orderedList",
+            "taskList",
+            "blockquote",
+          ]);
+          const matchedBlock = doc.content[idx];
+          if (matchedBlock && CONTAINER_TYPES.has(matchedBlock.type)) {
+            throw new Error(
+              `replaceText matched a ${matchedBlock.type} container block; replacing it would destroy the whole structure. ` +
+                `Use afterText to insert near it, or updatePageJson for surgical edits.`,
+            );
+          }
+          doc.content.splice(idx, 1, node);
+          placement = "replaced";
+        } else if (opts.afterText) {
+          // Ambiguity guard (mirrors editPageText): refuse a non-unique fragment.
+          const matches = doc.content.filter((b: any) =>
+            blockText(b).includes(opts.afterText!),
+          );
+          if (matches.length === 0) {
+            throw new Error(`afterText not found: "${opts.afterText}"`);
+          }
+          if (matches.length > 1) {
+            throw new Error(
+              `afterText "${opts.afterText}" matches ${matches.length} blocks; use a longer unique fragment`,
+            );
+          }
+          const idx = doc.content.findIndex((b: any) =>
+            blockText(b).includes(opts.afterText!),
+          );
+          doc.content.splice(idx + 1, 0, node);
+          placement = "after";
+        } else {
+          doc.content.push(node);
+          placement = "appended";
+        }
+
+        return doc;
+      },
+    );
+
+    return {
+      success: true,
+      pageId,
+      attachmentId: up.attachmentId,
+      src: up.src,
+      placement,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Replace an existing image in a page with a new image fetched from a web
+   * (http/https) URL. Uploads the new file as a brand-new attachment, which
+   * yields a fresh clean URL that both renders correctly and busts browser
+   * caches (the URL changed). Finds every image node
+   * whose attrs.attachmentId === oldAttachmentId (recursively, incl. nodes nested
+   * in callouts/tables) and repoints its src/attachmentId/size, preserving
+   * comments, alignment and alt. Operates on the live collab document so comments
+   * and concurrent edits are preserved. Throws if no matching image is found.
+   *
+   * The OLD attachment is left in place as an unreferenced orphan: Docmost
+   * exposes NO HTTP API to delete a single content attachment (verified against
+   * the attachment controller/service and by probing the live API — deletion
+   * happens only by cascade when the page, space or user is removed). This is the
+   * same outcome as Docmost's own editor when an image is removed/replaced.
+   * In-place byte overwrite is deliberately NOT used because some Docmost
+   * versions corrupt the attachment (HTTP 500) when its bytes are overwritten.
+   */
+  async replaceImage(
+    pageId: string,
+    oldAttachmentId: string,
+    url: string,
+    opts: { align?: "left" | "center" | "right"; alt?: string } = {},
+  ) {
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260). The
+    // page lock must ALSO key on the UUID so this operation serializes against
+    // other writes to the same page (mutatePageContent now locks by the resolved
+    // UUID too); locking by the raw slugId here would desync the mutex key and
+    // reopen the TOCTOU/orphan-attachment window the lock closes. uploadImage
+    // keeps the agent-supplied id (it hits REST, not the collab doc).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Hold ONE per-page lock for the WHOLE operation (scan -> upload -> write).
+    // Previously the scan and the write were two separate mutatePageContent
+    // calls, each acquiring + releasing the lock, with the upload happening in
+    // the UNLOCKED gap between them. A concurrent op could interleave there: it
+    // could remove the target image so the write pass matches nothing, leaving
+    // the freshly-uploaded attachment as an un-deletable orphan (Docmost has no
+    // API to delete a single content attachment). Acquiring the lock once and
+    // using the non-locking collab helper inside (the per-page mutex is NOT
+    // reentrant, so the self-locking mutatePageContent would deadlock here)
+    // closes that TOCTOU window. uploadImage hits /files/upload over plain HTTP
+    // and does not touch the page lock, so it is safe to call while held.
+    return withPageLock(pageUuid, async () => {
+      // STEP 1: read-only live check. Scan the live document for any image node
+      // matching oldAttachmentId BEFORE uploading anything, so a wrong/stale id
+      // throws without ever creating an orphan attachment.
+      let matchFound = false;
+      const scan = (nodes: any[]) => {
+        for (const node of nodes) {
+          if (!node) continue;
+          if (
+            node.type === "image" &&
+            node.attrs &&
+            node.attrs.attachmentId === oldAttachmentId
+          ) {
+            matchFound = true;
+          }
+          if (Array.isArray(node.content)) scan(node.content);
+        }
+      };
+
+      await this.mutateLiveContentUnlocked(pageUuid, collabToken, (liveDoc) => {
+        matchFound = false; // reset per-transform (collab may retry the read).
+        const doc =
+          liveDoc && liveDoc.type === "doc"
+            ? liveDoc
+            : { type: "doc", content: [] };
+        if (Array.isArray(doc.content)) scan(doc.content);
+        return null; // read-only: never write on the check pass.
+      });
+
+      if (!matchFound) {
+        throw new Error(
+          `replaceImage: no image with attachmentId "${oldAttachmentId}" found on page ${pageId}`,
+        );
+      }
+
+      // STEP 2: a match exists — upload the new file as a FRESH attachment (new
+      // id, new clean URL) and repoint every matching node in a second pass.
+      // Still inside the SAME lock, so no other op can have changed the page
+      // since the scan.
+      const up = await this.uploadImage(pageId, url);
+
+      let replaced = 0;
+
+      // Swap the source of one image node, preserving align/alt/title/geometry.
+      const repoint = (node: any) => {
+        node.attrs = {
+          ...node.attrs,
+          src: up.src,
+          attachmentId: up.attachmentId,
+          // Default to null when fileSize is unknown so the attr is never
+          // undefined.
+          size: up.fileSize ?? null,
+        };
+        if (opts.align) node.attrs.align = opts.align;
+        if (opts.alt !== undefined) node.attrs.alt = opts.alt;
+        replaced++;
+      };
+
+      // Recursively repoint every image node (incl. ones nested in callouts/tables).
+      const walk = (nodes: any[]) => {
+        for (const node of nodes) {
+          if (!node) continue;
+          if (
+            node.type === "image" &&
+            node.attrs &&
+            node.attrs.attachmentId === oldAttachmentId
+          ) {
+            repoint(node);
+          }
+          if (Array.isArray(node.content)) walk(node.content);
+        }
+      };
+
+      const mutation = await this.mutateLiveContentUnlocked(
+        pageUuid,
+        collabToken,
+        (liveDoc) => {
+          // Reset per-transform so collab retries recompute cleanly (no double-count).
+          replaced = 0;
+          const doc =
+            liveDoc && liveDoc.type === "doc"
+              ? liveDoc
+              : { type: "doc", content: [] };
+          if (!Array.isArray(doc.content)) doc.content = [];
+          walk(doc.content);
+          if (replaced === 0) return null; // no match -> skip the write entirely
+          return doc;
+        },
+      );
+      // KNOWN LIMITATION: a same-count image SRC swap (image count unchanged, no
+      // text/mark change) may still report verify.changed === false, because the
+      // text+marks+integrity-count model in summarizeChange does not inspect
+      // image `src`/attachmentId attributes. That is acceptable here — the
+      // replace is confirmed by `replaced` below, and verify is supplementary.
+
+      if (replaced === 0) {
+        // The pass-1 SCAN found the target (matchFound was true) and we already
+        // uploaded the new attachment, but pass-2 matched nothing — a concurrent
+        // editor must have removed the node between the two passes. Do NOT throw
+        // here (that would leak the just-uploaded attachment AND report failure);
+        // instead report success with the upload flagged as an unreferenced
+        // orphan so the caller knows. (The early throw above still covers the
+        // case where pass-1 finds nothing, before any upload happens.)
+        return {
+          success: true,
+          replaced: 0,
+          pageId,
+          oldAttachmentId,
+          newAttachmentId: up.attachmentId,
+          src: up.src,
+          orphanedAttachmentId: up.attachmentId,
+          warning:
+            "target image was removed concurrently; uploaded attachment is unreferenced",
+          verify: mutation.verify,
+        };
+      }
+
+      return {
+        success: true,
+        pageId,
+        replaced,
+        oldAttachmentId,
+        newAttachmentId: up.attachmentId,
+        src: up.src,
+        verify: mutation.verify,
+      };
+    });
+  }
+
+  // --- draw.io diagrams (issue #423) ---
+
+  /**
+   * Upload a ready-made byte buffer as a page attachment via the same
+   * multipart /files/upload endpoint uploadImage uses. Split out as its own
+   * (overridable) seam so drawioCreate/update can upload the generated
+   * `.drawio.svg` without going through the URL-fetch path, and so tests can
+   * stub the network. Mirrors uploadImage's fresh-FormData + one-shot 401/403
+   * re-auth handling (a FormData body is single-use, so it must be rebuilt per
+   * attempt).
+   */
+
+  // --- draw.io diagrams (issue #423) ---
+
+  /**
+   * Upload a ready-made byte buffer as a page attachment via the same
+   * multipart /files/upload endpoint uploadImage uses. Split out as its own
+   * (overridable) seam so drawioCreate/update can upload the generated
+   * `.drawio.svg` without going through the URL-fetch path, and so tests can
+   * stub the network. Mirrors uploadImage's fresh-FormData + one-shot 401/403
+   * re-auth handling (a FormData body is single-use, so it must be rebuilt per
+   * attempt).
+   */
+  protected async uploadAttachmentBuffer(
+    pageId: string,
+    buffer: Buffer,
+    fileName: string,
+    mime: string,
+  ): Promise<{ id: string; fileName: string; fileSize: number }> {
+    await this.ensureAuthenticated();
+    const buildForm = () => {
+      const form = new FormData();
+      form.append("pageId", pageId);
+      form.append("file", buffer, { filename: fileName, contentType: mime });
+      return form;
+    };
+    const uploadUrl = `${this.apiUrl}/files/upload`;
+    let response;
+    try {
+      const form = buildForm();
+      response = await axios.post(uploadUrl, form, {
+        headers: {
+          ...form.getHeaders(),
+          Authorization: this.client.defaults.headers.common["Authorization"],
+        },
+        timeout: 60000,
+      });
+    } catch (error) {
+      if (
+        axios.isAxiosError(error) &&
+        (error.response?.status === 401 || error.response?.status === 403)
+      ) {
+        await this.login();
+        const form2 = buildForm();
+        response = await axios.post(uploadUrl, form2, {
+          headers: {
+            ...form2.getHeaders(),
+            Authorization: this.client.defaults.headers.common["Authorization"],
+          },
+          timeout: 60000,
+        });
+      } else if (axios.isAxiosError(error)) {
+        if (process.env.DEBUG) {
+          console.error(
+            "Attachment upload failed; response body:",
+            JSON.stringify(error.response?.data),
+          );
+        }
+        throw new Error(
+          `Attachment upload failed: ${error.response?.status} ${error.response?.statusText}`,
+        );
+      } else {
+        throw error;
+      }
+    }
+    const att = response.data?.data ?? response.data;
+    if (!att?.id || !att?.fileName) {
+      throw new Error(
+        "Unexpected /files/upload response: " + JSON.stringify(response.data),
+      );
+    }
+    return {
+      id: att.id,
+      fileName: att.fileName,
+      fileSize: att.fileSize ?? buffer.length,
+    };
+  }
+
+  /**
+   * Fetch a stored `.drawio.svg` attachment as text. Overridable seam over
+   * fetchInternalFile (the authed loopback fetch, which also rejects any
+   * traversal/SSRF src) so drawioGet/update can read the current diagram and
+   * tests can stub the bytes.
+   */
+  protected async fetchAttachmentText(src: string): Promise<string> {
+    const { buffer } = await this.fetchInternalFile(src);
+    return buffer.toString("utf-8");
+  }
+
+  /**
+   * Resolve a drawio node on a page by `attrs.id` or `#<index>` and return the
+   * node plus its ref. Throws a clear error if the ref does not resolve to a
+   * drawio node.
+   */
+  }
+  return MediaMixin;
+}

packages/mcp/src/client/nodes-write.ts

@@ -0,0 +1,700 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import {
+  updatePageContentRealtime,
+  replacePageContent,
+  markdownToProseMirror,
+  markdownToProseMirrorCanonical,
+  mutatePageContent,
+  assertYjsEncodable,
+  MutationResult,
+} from "../lib/collaboration.js";
+import {
+  replaceNodeById,
+  replaceNodeByIdWithMany,
+  reassignCollidingBlockIds,
+  deleteNodeById,
+  assertUnambiguousMatch,
+  insertNodeRelative,
+  insertNodesRelative,
+  blockPlainText,
+  buildOutline,
+  getNodeByRef,
+  readTable,
+  insertTableRow,
+  deleteTableRow,
+  updateTableCell,
+  findInvalidNode,
+} from "@docmost/prosemirror-markdown";
+import {
+  importMarkdownFragment,
+  canBeDocChild,
+  findUnrepresentableTableAttrs,
+} from "../lib/markdown-fragment.js";
+import {
+  applyTextEdits,
+  TextEdit,
+  TextEditResult,
+  TextEditFailure,
+} from "../lib/json-edit.js";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+import { normalizeAndMergeFootnotes } from "../lib/footnote-normalize-merge.js";
+
+// Public method surface of NodesWriteMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements INodesWriteMixin` fails to compile on drift.
+export interface INodesWriteMixin {
+  updatePageJson(pageId: string, doc?: any, title?: string): any;
+  editPageText(pageId: string, edits: TextEdit[]): any;
+  patchNode(pageId: string, nodeId: string, input: { markdown?: string; node?: any }): any;
+  insertNode(pageId: string, input: { markdown?: string; node?: any }, opts: { position: "before" | "after" | "append"; anchorNodeId?: string; anchorText?: string; }): any;
+  deleteNode(pageId: string, nodeId: string): any;
+}
+
+export function NodesWriteMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & INodesWriteMixin> & TBase {
+  abstract class NodesWriteMixin extends Base implements INodesWriteMixin {
+  /**
+   * Replace page content with a raw ProseMirror JSON document (lossless) and/or
+   * update its title. Both `doc` and `title` are optional, but at least one must
+   * be supplied:
+   *  - `doc` provided   -> validate + full-overwrite the body (and update the
+   *                        title too when `title` is also given).
+   *  - `doc` omitted, `title` given -> title-only update; the body is NOT
+   *                        touched/resent (no collab write happens).
+   *  - neither given    -> throws (nothing to update).
+   */
+  async updatePageJson(pageId: string, doc?: any, title?: string) {
+    await this.ensureAuthenticated();
+
+    // Title-only / no-op handling: when no document is supplied, do NOT write
+    // the body. Update the title if one was given; otherwise there is nothing
+    // to do, so fail loudly rather than silently no-op.
+    if (doc == null) {
+      if (!title) {
+        throw new Error(
+          "updatePageJson: nothing to update (provide content and/or title)",
+        );
+      }
+      await this.client.post("/pages/update", { pageId, title });
+      return {
+        success: true,
+        modified: true,
+        message: "Page title updated (content left unchanged).",
+        pageId,
+      };
+    }
+
+    // Validate the document shape before a full overwrite: a malformed doc
+    // would otherwise silently corrupt the page (full-overwrite is the
+    // documented behaviour; no optimistic-concurrency is applied here).
+    if (
+      typeof doc !== "object" ||
+      doc.type !== "doc" ||
+      !Array.isArray(doc.content)
+    ) {
+      throw new Error(
+        'content must be a ProseMirror document ({"type":"doc","content":[...]}) ' +
+          "where content is an array of nodes each having a string `type`",
+      );
+    }
+
+    // Recurse the WHOLE document so a malformed nested node (e.g. a node with a
+    // non-string type, a non-array content/marks, or a text node missing its
+    // string text) is rejected up front rather than silently corrupting the
+    // page on overwrite.
+    this.validateDocStructure(doc);
+
+    // #409: beyond the string-`type` check above, reject a nested node whose
+    // `type` is a string but NOT a known Docmost schema node (a typo/unknown
+    // block) — the same `Unknown node type` the encoder throws — with a rich,
+    // path-anchored message, still BEFORE any collab connection.
+    this.assertValidNodeShape("updatePageJson", doc);
+
+    // Sanitize URLs before writing. This closes the JSON-path bypass: unlike
+    // the markdown link path (which TipTap sanitizes), raw JSON could otherwise
+    // inject javascript:/data: link hrefs or media srcs straight into the doc.
+    this.validateDocUrls(doc);
+
+    // Canonicalize footnotes (idempotent): an agent-authored JSON doc cannot
+    // leave footnotes out of order, orphaned, or in multiple lists — the bottom
+    // list + numbering are always derived from reference order. No-op when the
+    // footnotes are already canonical.
+    // #419: normalize + merge glyph-forked definitions before canonicalizing.
+    doc = normalizeAndMergeFootnotes(doc);
+    doc = canonicalizeFootnotes(doc);
+
+    // Write the BODY first, then the title (#159 split-brain): a failed body
+    // write (e.g. persist timeout) must not leave a new title over the old body.
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+    const mutation = await this.replacePage(
+      pageUuid,
+      doc,
+      collabToken,
+      this.apiUrl,
+    );
+
+    // Body persisted successfully — now it is safe to set the title.
+    if (title) {
+      await this.client.post("/pages/update", { pageId, title });
+    }
+
+    return {
+      success: true,
+      modified: true,
+      message: "Page content replaced from ProseMirror JSON.",
+      pageId,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * AUTHOR-INLINE footnote insertion. The agent supplies only WHERE
+   * (`anchorText`, a snippet of body text to attach the marker after) and WHAT
+   * (`text`, the footnote content as markdown). Numbering and the bottom
+   * `footnotesList` are derived deterministically server-side
+   * (`insertInlineFootnote` -> `canonicalizeFootnotes`): the agent never sees,
+   * assigns, or edits a footnote number or the list, so it CANNOT desync.
+   *
+   * Content DEDUP: when an existing definition has the same content, its id is
+   * reused (one number, one definition, several references). The write is atomic
+   * via `mutatePageContent` (single-writer, page-locked); if the anchor text is
+   * not found the transform aborts with a clear error and no write happens.
+   */
+
+  /**
+   * Surgical text edits: find/replace inside text nodes of the live
+   * document. Preserves all block ids, marks, callouts and tables.
+   */
+  async editPageText(pageId: string, edits: TextEdit[]) {
+    await this.ensureAuthenticated();
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Apply the edits against the LIVE synced document, not the debounced REST
+    // snapshot, so concurrent human edits/comments are preserved. applyTextEdits
+    // records per-edit match problems in `failed` instead of throwing, and
+    // applies whatever it can; we abort the write only when nothing applied.
+    let results: TextEditResult[] | undefined;
+    let failed: TextEditFailure[] | undefined;
+    // Whether we actually wrote new content. Set inside the transform: a
+    // degenerate edit (e.g. find === replace, or a batch that nets to no change)
+    // can "apply" yet leave the document byte-for-byte identical, in which case
+    // we must NOT write (no spurious history version) and must not claim a write
+    // happened.
+    let wrote = false;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        wrote = false;
+        const r = applyTextEdits(liveDoc, edits);
+        results = r.results;
+        failed = r.failed;
+        // Nothing applied -> abort the write (mutatePageContent treats a null
+        // return from the transform as "write nothing").
+        if (r.results.length === 0) return null;
+        // Edits "applied" but produced an identical document: skip the write so
+        // no new history version is created. Stable structural comparison via
+        // JSON.stringify (both docs come from the same deep-copied source, so
+        // key order is stable).
+        if (JSON.stringify(r.doc) === JSON.stringify(liveDoc)) return null;
+        wrote = true;
+        return r.doc;
+      },
+    );
+
+    if ((results?.length ?? 0) === 0 && (failed?.length ?? 0) > 0) {
+      // No edit applied: surface an aggregated, actionable error so the caller
+      // does not mistake a no-op for a partial success.
+      throw new Error(
+        "editPageText: no edits were applied (nothing written). " +
+          failed!.map((f) => `"${f.find}": ${f.reason}`).join("; "),
+      );
+    }
+
+    // Edits matched but produced no content change (identical document): report
+    // a successful no-op — NOT a failure — and do not falsely claim a write.
+    if (!wrote) {
+      return {
+        success: true,
+        pageId,
+        applied: results,
+        failed,
+        message: "No changes written (edits produced identical content).",
+        verify: mutation.verify,
+      };
+    }
+
+    const result: any = {
+      success: true,
+      pageId,
+      applied: results,
+      failed,
+      message:
+        (failed?.length ?? 0)
+          ? `Applied ${results?.length ?? 0} edit(s); ${failed!.length} failed (see failed[]). Node ids and formatting preserved.`
+          : "Text edits applied (node ids and formatting preserved).",
+      verify: mutation.verify,
+    };
+
+    // If any applied edit matched only after stripping markdown (the
+    // normalized fallback), warn that editPageText preserved existing marks
+    // and did NOT change formatting — so a caller who intended a formatting
+    // change is pointed at patchNode.
+    if (results?.some((r) => r.normalized === true)) {
+      result.warning =
+        "Some edits matched only after stripping markdown from your find string; " +
+        "editPageText preserved existing marks (it did not change bold/strike/etc.). " +
+        "If you intended a formatting change, use patchNode.";
+    }
+
+    return result;
+  }
+
+  /**
+   * Replace the block whose attrs.id === nodeId. Operates on the LIVE collab
+   * document so comments and concurrent edits are preserved.
+   *
+   * Exactly one of `input.markdown` / `input.node` (#413):
+   *  - `markdown` (RECOMMENDED): the block is rewritten from a canonical markdown
+   *    fragment. The fragment may import to N blocks (a "1 -> N" splice: rewrite a
+   *    whole section in one call). The FIRST resulting block INHERITS the target's
+   *    `attrs.id` (so an existing comment anchoring the block by id survives); the
+   *    rest get FRESH ids. `^[...]` footnotes in the fragment are first-class:
+   *    their definitions merge into the page's TAIL footnote list (content-key
+   *    dedup + canonicalize), same machinery insertFootnote uses. REJECTED when
+   *    the TARGET block carries a table-cell attribute markdown cannot represent
+   *    (colspan/rowspan/colwidth/background) — use the table tools or `node`.
+   *  - `node`: a raw ProseMirror node for precise attr/mark work. The replacement
+   *    keeps the target id (if `node.attrs.id` is missing it is set to nodeId).
+   *
+   * #159 ambiguous-id semantics are unchanged: 0 matches -> "no node"; >1 matches
+   * -> "ambiguous, refused" (nothing written), on BOTH paths — the markdown path
+   * runs a dry `replaceNodeById` count first, so a duplicated id never splices.
+   */
+  async patchNode(
+    pageId: string,
+    nodeId: string,
+    input: { markdown?: string; node?: any },
+  ) {
+    await this.ensureAuthenticated();
+
+    // XOR: exactly one of markdown / node. Both optional in the schema; the
+    // runtime enforces the recommendation ("markdown for prose, node for fine
+    // work") without letting an ambiguous both-or-neither call through.
+    const hasMd =
+      input != null &&
+      typeof input.markdown === "string" &&
+      input.markdown.trim() !== "";
+    const hasNode = input != null && input.node != null;
+    if (hasMd === hasNode) {
+      throw new Error(
+        "patchNode: provide exactly one of `markdown` (recommended, for prose) " +
+          "or `node` (a raw ProseMirror node, for precise attr/mark work)",
+      );
+    }
+
+    if (hasMd) {
+      return this.patchNodeMarkdown(pageId, nodeId, input.markdown as string);
+    }
+    return this.patchNodeJson(pageId, nodeId, input.node);
+  }
+
+  /**
+   * patchNode with a raw ProseMirror `node` (the pre-#413 behavior). Replaces
+   * EVERY node whose attrs.id === nodeId; the swapped-in node keeps the target
+   * id. #159 ambiguity refused. Split out so the markdown path can reuse the
+   * shared collab/guard plumbing without a giant branch.
+   */
+  protected async patchNodeJson(pageId: string, nodeId: string, node: any) {
+    if (!node || typeof node !== "object" || typeof node.type !== "string") {
+      throw new Error(
+        "patchNode: `node` must be an object with a string `type`",
+      );
+    }
+    // Preserve the block id WITHOUT mutating the caller's object: build a local
+    // copy whose attrs.id === nodeId (so the swapped-in node keeps the id of the
+    // node it replaces).
+    const target = {
+      ...node,
+      attrs: {
+        ...(node.attrs && typeof node.attrs === "object" ? node.attrs : {}),
+      },
+    };
+    if (target.attrs.id == null) {
+      target.attrs.id = nodeId;
+    }
+
+    // #409: fail fast on a malformed node SHAPE (a nested child with an
+    // absent/unknown `type`, e.g. a text leaf written as `{"text":"foo"}` with
+    // no `"type":"text"`) BEFORE opening a collab session or taking the page
+    // lock — the root-only `typeof node.type === "string"` check above never
+    // sees nested children, and the encoder's `Unknown node type: undefined`
+    // would otherwise only surface after the connection.
+    this.assertValidNodeShape("patchNode", target);
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Track the replacement count in an outer var, reset per-transform, so a
+    // collab retry recomputes it cleanly (mirrors replaceImage's pattern).
+    let replaced = 0;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        replaced = 0;
+        const { doc: nd, replaced: r } = replaceNodeById(
+          liveDoc,
+          nodeId,
+          target,
+        );
+        replaced = r;
+        // 0 matches -> skip the write. >1 matches -> the id is AMBIGUOUS: Docmost
+        // duplicates block ids on copy/paste (and copyPageContent writes them
+        // verbatim), so replacing "the node with id X" would silently clobber
+        // EVERY duplicate (#159). Refuse: skip the write and throw below so the
+        // model re-targets with a more specific anchor instead of corrupting the
+        // page. Only an unambiguous single match is written.
+        if (replaced !== 1) return null;
+        return nd;
+      },
+    );
+
+    // 0 -> "no node"; >1 -> "ambiguous, refused" (the transform already skipped
+    // the write for any count !== 1). Single shared guard (#159, #185 review).
+    assertUnambiguousMatch("patchNode", "replace", replaced, nodeId, pageId);
+
+    return { success: true, replaced, nodeId, verify: mutation.verify };
+  }
+
+  /**
+   * patchNode with a MARKDOWN fragment (#413). Imports the fragment through the
+   * canonical importer, then 1 -> N splices the resulting blocks in place of the
+   * target block on the LIVE collab doc:
+   *  - the FIRST block inherits the target's id; the rest get FRESH ids (minted
+   *    by the importer/id-remap, so neighbour blocks are untouched);
+   *  - `^[...]` footnote definitions merge into the page's tail list;
+   *  - REJECTED when the target block carries a markdown-unrepresentable table
+   *    attr (colspan/rowspan/colwidth/background) — guarding against silent loss;
+   *  - #159 ambiguity is enforced by a dry `replaceNodeById` count BEFORE the
+   *    splice, so a duplicated id never writes.
+   */
+  protected async patchNodeMarkdown(
+    pageId: string,
+    nodeId: string,
+    markdown: string,
+  ) {
+    // Import the fragment up front (network-free, canonical) so a bad fragment
+    // fails before any collab connection or page lock.
+    const { blocks, definitions } = await importMarkdownFragment(markdown);
+
+    // The first imported block inherits the target id; the rest keep the fresh
+    // ids the importer assigned. Build the thread now so it is stable across a
+    // collab retry (the transform below is pure over its inputs).
+    const threaded = blocks.map((b, i) => {
+      if (i !== 0) return b;
+      return {
+        ...b,
+        attrs: {
+          ...(b && typeof b.attrs === "object" ? b.attrs : {}),
+          id: nodeId,
+        },
+      };
+    });
+
+    // Shape-validate every imported block up front (parity with the JSON path):
+    // the importer only emits schema nodes, but the check is cheap insurance and
+    // yields the same rich #409 diagnostics if the schema ever drifts.
+    for (const b of threaded) {
+      this.assertValidNodeShape("patchNode", b);
+    }
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    let replaced = 0;
+    let guardAttrs: string | null = null;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        replaced = 0;
+        guardAttrs = null;
+
+        // #159: count matches with the same recursive walk the JSON path uses;
+        // only an UNAMBIGUOUS single match may write. A dry count keeps the
+        // ambiguity semantics identical across both paths.
+        const { replaced: count } = replaceNodeById(liveDoc, nodeId, {
+          type: "paragraph",
+        });
+        replaced = count;
+        if (count !== 1) return null;
+
+        // Guard against SILENT LOSS: if the target block carries a table-cell
+        // attribute markdown cannot represent (colspan/rowspan/colwidth/
+        // background), refuse the markdown rewrite so those attrs are not
+        // dropped. Simple tables (no such attrs) rewrite fine.
+        const hit = getNodeByRef(liveDoc, nodeId);
+        guardAttrs = hit ? findUnrepresentableTableAttrs(hit.node) : null;
+        if (guardAttrs != null) return null;
+
+        // Re-mint any minted block id that collides with an existing page id
+        // (skip index 0: its id is intentionally the target nodeId, unique by
+        // the #159 dry-count above), so the 1 -> N splice stays page-wide unique.
+        reassignCollidingBlockIds(liveDoc, threaded, 0);
+
+        // 1 -> N splice, then merge any fragment footnote definitions into the
+        // page's tail list and re-derive canonical footnote numbering.
+        const { doc: spliced } = replaceNodeByIdWithMany(
+          liveDoc,
+          nodeId,
+          threaded,
+        );
+        return mergeFootnoteDefinitions(spliced, definitions);
+      },
+    );
+
+    // Surface the guard rejection with an actionable message (nothing written).
+    if (guardAttrs != null) {
+      throw new Error(
+        `patchNode: the target block has table-cell attributes markdown cannot ` +
+          `represent (${guardAttrs}) — a markdown rewrite would drop them. Use ` +
+          `the table tools (tableUpdateCell/tableInsertRow) or pass a raw ` +
+          `ProseMirror \`node\` instead of \`markdown\`.`,
+      );
+    }
+
+    // 0 -> "no node"; >1 -> "ambiguous, refused" (the transform skipped the write
+    // for any count !== 1). Shared #159 guard, identical to the JSON path.
+    assertUnambiguousMatch("patchNode", "replace", replaced, nodeId, pageId);
+
+    return {
+      success: true,
+      replaced,
+      nodeId,
+      blocks: threaded.length,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Insert content relative to an anchor (or append it at the top level).
+   * Operates on the LIVE collab document so comments and concurrent edits are
+   * preserved.
+   *
+   * Exactly one of `input.markdown` / `input.node` (#413):
+   *  - `markdown` (RECOMMENDED): a canonical markdown fragment. It may import to
+   *    SEVERAL blocks — they are inserted IN ORDER at the anchor. `^[...]`
+   *    footnote definitions merge into the page's tail list (same machinery as
+   *    insertFootnote). Every inserted block gets a fresh id.
+   *  - `node`: a raw ProseMirror node for precise attr/mark work, or to insert
+   *    table structure (a bare tableRow/tableCell/tableHeader — NOT expressible in
+   *    markdown, so those stay JSON-only).
+   *
+   * opts.position:
+   *  - "append": push the content at the end of the top-level content.
+   *  - "before"/"after": insert as a sibling of the anchor, just before/after it.
+   *    Exactly one of anchorNodeId / anchorText must be given; anchorNodeId
+   *    locates a node anywhere by attrs.id, anchorText matches the first top-level
+   *    block whose plain text includes it.
+   *
+   * Throws if the anchor cannot be found.
+   */
+  async insertNode(
+    pageId: string,
+    input: { markdown?: string; node?: any },
+    opts: {
+      position: "before" | "after" | "append";
+      anchorNodeId?: string;
+      anchorText?: string;
+    },
+  ) {
+    await this.ensureAuthenticated();
+
+    // XOR: exactly one of markdown / node (both optional in the schema).
+    const hasMd =
+      input != null &&
+      typeof input.markdown === "string" &&
+      input.markdown.trim() !== "";
+    const hasNode = input != null && input.node != null;
+    if (hasMd === hasNode) {
+      throw new Error(
+        "insertNode: provide exactly one of `markdown` (recommended, for prose) " +
+          "or `node` (a raw ProseMirror node, for precise attr/mark work or table structure)",
+      );
+    }
+
+    if (
+      !opts ||
+      (opts.position !== "before" &&
+        opts.position !== "after" &&
+        opts.position !== "append")
+    ) {
+      throw new Error(
+        'insertNode: `position` must be one of "before", "after", "append"',
+      );
+    }
+    if (opts.position === "before" || opts.position === "after") {
+      // before/after require EXACTLY ONE anchor (an id or a text fragment).
+      const hasId =
+        typeof opts.anchorNodeId === "string" && opts.anchorNodeId.length > 0;
+      const hasText =
+        typeof opts.anchorText === "string" && opts.anchorText.length > 0;
+      if (hasId === hasText) {
+        throw new Error(
+          `insertNode: position "${opts.position}" requires exactly one of anchorNodeId or anchorText`,
+        );
+      }
+    }
+
+    // Resolve the ordered list of blocks to insert plus any footnote definitions
+    // to merge. The markdown path imports canonically (so an inserted block is
+    // byte-identical to the same content in a full-page import); the node path is
+    // a single block with no footnote merge (raw JSON `^[...]` is not touched).
+    let blocks: any[];
+    let definitions: any[] = [];
+    if (hasMd) {
+      const frag = await importMarkdownFragment(input.markdown as string);
+      blocks = frag.blocks;
+      definitions = frag.definitions;
+    } else {
+      const node = input.node;
+      if (!node || typeof node !== "object" || typeof node.type !== "string") {
+        throw new Error(
+          "insertNode: `node` must be an object with a string `type`",
+        );
+      }
+      blocks = [node];
+    }
+
+    // #409: fail fast on a malformed node SHAPE (a nested child with an
+    // absent/unknown `type`) BEFORE opening a collab session or taking the page
+    // lock — the root-only check above never sees nested children.
+    for (const b of blocks) {
+      this.assertValidNodeShape("insertNode", b);
+    }
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Track insertion in an outer var, reset per-transform, so a collab retry
+    // recomputes it cleanly (mirrors replaceImage's pattern).
+    let inserted = false;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        inserted = false;
+        // Re-mint any minted block id that collides with an existing page id
+        // (all inserted blocks are fresh, no skip) so the splice stays unique.
+        if (hasMd) reassignCollidingBlockIds(liveDoc, blocks);
+        // Single-block node path keeps `insertNodeRelative` (it owns the
+        // structural table-node splicing); the markdown path uses the array
+        // splice so N blocks land in order at one anchor.
+        const res = hasMd
+          ? insertNodesRelative(liveDoc, blocks, opts)
+          : insertNodeRelative(liveDoc, blocks[0], opts);
+        inserted = res.inserted;
+        if (!inserted) return null; // anchor not found -> skip the write entirely
+        // Merge any fragment footnote definitions into the page tail list and
+        // re-derive canonical numbering (no-op when there are none).
+        return mergeFootnoteDefinitions(res.doc, definitions);
+      },
+    );
+
+    if (!inserted) {
+      const anchorDesc = opts.anchorNodeId
+        ? `anchorNodeId "${opts.anchorNodeId}"`
+        : `anchorText "${opts.anchorText}"`;
+      // anchorText is matched against the block's literal RENDERED plain text;
+      // markdown/emoji are tolerated only as a strip-and-retry fallback, so a
+      // miss usually means the text differs from what's on the page.
+      const hint = opts.anchorText
+        ? " anchorText must be the block's literal rendered plain text (no markdown wrappers or emoji); anchorNodeId from getPageJson is more reliable."
+        : "";
+      throw new Error(
+        `insertNode: anchor not found (${anchorDesc}) on page ${pageId}.${hint}`,
+      );
+    }
+
+    return {
+      success: true,
+      inserted: true,
+      position: opts.position,
+      blocks: blocks.length,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Remove EVERY node whose attrs.id === nodeId (recursively, including nodes
+   * nested in callouts/tables) from its parent content array. Operates on the
+   * LIVE collab document so comments and concurrent edits are preserved.
+   * Throws if no node matches.
+   */
+  async deleteNode(pageId: string, nodeId: string) {
+    await this.ensureAuthenticated();
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Track the deletion count in an outer var, reset per-transform, so a
+    // collab retry recomputes it cleanly (mirrors replaceImage's pattern).
+    let deleted = 0;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        deleted = 0;
+        const { doc: nd, deleted: d } = deleteNodeById(liveDoc, nodeId);
+        deleted = d;
+        // 0 matches -> skip the write. >1 matches -> the id is AMBIGUOUS (block
+        // ids are duplicated on copy/paste, #159): deleting "the node with id X"
+        // would silently remove EVERY duplicate. Refuse: skip the write and throw
+        // below so the model re-targets. Only an unambiguous single match is
+        // deleted.
+        if (deleted !== 1) return null;
+        return nd;
+      },
+    );
+
+    // 0 -> "no node"; >1 -> "ambiguous, refused" (the transform already skipped
+    // the write for any count !== 1). Single shared guard (#159, #185 review).
+    assertUnambiguousMatch("deleteNode", "delete", deleted, nodeId, pageId);
+
+    return { success: true, deleted, nodeId, verify: mutation.verify };
+  }
+
+  /** Build the public share URL for a page. */
+  }
+  return NodesWriteMixin;
+}

packages/mcp/src/client/pages.ts

@@ -0,0 +1,655 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import FormData from "form-data";
+import axios, { AxiosInstance } from "axios";
+import { convertProseMirrorToMarkdown } from "../lib/markdown-converter.js";
+import {
+  updatePageContentRealtime,
+  replacePageContent,
+  markdownToProseMirror,
+  markdownToProseMirrorCanonical,
+  mutatePageContent,
+  assertYjsEncodable,
+  MutationResult,
+} from "../lib/collaboration.js";
+import { footnoteWarningsField } from "../lib/footnote-analyze.js";
+import {
+  serializeDocmostMarkdown,
+  parseDocmostMarkdown,
+} from "../lib/markdown-document.js";
+import { diffDocs, summarizeChange } from "../lib/diff.js";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+import { normalizeAndMergeFootnotes } from "../lib/footnote-normalize-merge.js";
+import vm from "node:vm";
+
+// Public method surface of PagesMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements IPagesMixin` fails to compile on drift.
+export interface IPagesMixin {
+  createPage(title: string, content: string, spaceId: string, parentPageId?: string): any;
+  updatePage(pageId: string, content: string, title?: string): any;
+  renamePage(pageId: string, title: string): any;
+  movePage(pageId: string, parentPageId: string | null, position?: string): any;
+  deletePage(pageId: string): any;
+  sharePage(pageId: string, searchIndexing?: boolean): any;
+  listShares(): any;
+  unsharePage(pageId: string): any;
+  exportPageMarkdown(pageId: string): Promise<string>;
+  importPageMarkdown(pageId: string, fullMarkdown: string): Promise<any>;
+  copyPageContent(sourcePageId: string, targetPageId: string): any;
+  listPageHistory(pageId: string, cursor?: string): any;
+  getPageHistory(historyId: string): any;
+  restorePageVersion(historyId: string): any;
+  diffPageVersions(pageId: string, from?: string, to?: string): any;
+}
+
+export function PagesMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & IPagesMixin> & TBase {
+  abstract class PagesMixin extends Base implements IPagesMixin {
+  /**
+   * Create a new page with title and content.
+   * Uses the /pages/import workaround (the only endpoint accepting content),
+   * then moves the page and restores the exact title: the import endpoint
+   * derives the title from the FILENAME and replaces spaces with
+   * underscores, so we explicitly re-set it via /pages/update afterwards.
+   */
+  async createPage(
+    title: string,
+    content: string,
+    spaceId: string,
+    parentPageId?: string,
+  ) {
+    await this.ensureAuthenticated();
+
+    if (parentPageId) {
+      try {
+        await this.getPage(parentPageId);
+      } catch (e) {
+        throw new Error(`Parent page with ID ${parentPageId} not found.`);
+      }
+    }
+
+    // 1. Create content via Import (using multipart/form-data).
+    // Build a FRESH FormData per send attempt: a FormData body is a single-use
+    // stream consumed on the first send, so it cannot be replayed by
+    // this.client's response interceptor (replay fails with 'socket hang up').
+    // Multipart re-auth is therefore done here with bare axios and an explicit
+    // one-shot 401/403 retry that rebuilds the body.
+    const fileContent = Buffer.from(content, "utf-8");
+    const buildForm = () => {
+      const form = new FormData();
+      form.append("spaceId", spaceId);
+      form.append("file", fileContent, {
+        filename: `${title || "import"}.md`,
+        contentType: "text/markdown",
+      });
+      return form;
+    };
+
+    const importUrl = `${this.apiUrl}/pages/import`;
+    let response;
+    try {
+      // Call buildForm() ONCE per attempt and reuse the instance for both
+      // getHeaders() and the body so the Content-Type boundary matches the body.
+      const form = buildForm();
+      // Read the Authorization header from this.client's defaults (set by
+      // login(), only ever deleted — never set to null) instead of building
+      // `Bearer ${this.token}`: a concurrent JSON 401 can null this.token
+      // mid-flight, which would otherwise produce a literal "Bearer null".
+      // ensureAuthenticated() above guarantees login() ran, so the default
+      // header exists here.
+      response = await axios.post(importUrl, form, {
+        headers: {
+          ...form.getHeaders(),
+          Authorization: this.client.defaults.headers.common["Authorization"],
+        },
+        timeout: 60000,
+      });
+    } catch (error) {
+      // On an expired-token auth error, re-login and retry exactly once with a
+      // freshly-rebuilt FormData (the previous one was already consumed).
+      if (
+        axios.isAxiosError(error) &&
+        (error.response?.status === 401 || error.response?.status === 403)
+      ) {
+        await this.login();
+        const form2 = buildForm();
+        response = await axios.post(importUrl, form2, {
+          headers: {
+            ...form2.getHeaders(),
+            Authorization: this.client.defaults.headers.common["Authorization"],
+          },
+          timeout: 60000,
+        });
+      } else {
+        throw error;
+      }
+    }
+    const newPageId = (response.data?.data ?? response.data).id;
+
+    // 2. Move to parent if needed
+    if (parentPageId) {
+      await this.movePage(newPageId, parentPageId);
+    }
+
+    // 3. Restore the exact title (import mangles spaces into underscores)
+    if (title) {
+      await this.client.post("/pages/update", { pageId: newPageId, title });
+    }
+
+    const page = await this.getPage(newPageId);
+    // Surface non-fatal footnote problems (dangling refs, empty/duplicate
+    // definitions, markers in tables) so the agent can fix its markup (#166).
+    return { ...page, ...footnoteWarningsField(content) };
+  }
+
+  /**
+   * Update a page's content from markdown and optionally its title.
+   * NOTE: full re-import — block ids regenerate. For surgical changes
+   * use editPageText / updatePageJson instead.
+   */
+  async updatePage(pageId: string, content: string, title?: string) {
+    await this.ensureAuthenticated();
+    // Open the collab doc by the canonical UUID, never the slugId (#260). The
+    // REST /pages/update title write below keeps the agent-supplied id (the
+    // server resolves a slugId there).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Write the BODY first, then the title (#159 split-brain). If the collab
+    // body write fails (e.g. a persist timeout), the title must be left
+    // UNTOUCHED so the page never ends up with a new title over its old body.
+    // A title write failing AFTER a successful body is rarer (REST is fast) and
+    // leaves correct content under a stale title — the lesser inconsistency.
+    let collabToken = "";
+    let mutation;
+    try {
+      collabToken = await this.getCollabTokenWithReauth();
+      mutation = await updatePageContentRealtime(
+        pageUuid,
+        content,
+        collabToken,
+        this.apiUrl,
+      );
+    } catch (error: any) {
+      // Verbose diagnostics (incl. anything that could expose a token prefix)
+      // are gated behind DEBUG; the thrown Error below carries no token data.
+      if (process.env.DEBUG) {
+        console.error(
+          "Failed to update page content via realtime collaboration:",
+          error,
+        );
+        const tokenPreview = collabToken
+          ? collabToken.substring(0, 15) + "..."
+          : "null";
+        console.error(`Collab token preview: ${tokenPreview}`);
+      }
+      throw new Error(`Failed to update page content: ${error.message}`);
+    }
+
+    // Body persisted successfully — now it is safe to set the title.
+    if (title) {
+      await this.client.post("/pages/update", { pageId, title });
+    }
+
+    return {
+      success: true,
+      modified: true,
+      message: "Page updated successfully.",
+      pageId: pageId,
+      verify: mutation.verify,
+      // Non-fatal footnote diagnostics (#166); omitted when there are none.
+      ...footnoteWarningsField(content),
+    };
+  }
+
+  /**
+   * Validate a URL string against a scheme allowlist for a given context.
+   *
+   * The markdown link path enforces safe schemes via TipTap, but the raw
+   * JSON path (updatePageJson) bypasses that — so this is the sanitization
+   * choke point for ProseMirror JSON written directly by the caller.
+   *
+   * - "link":  reject javascript:, vbscript:, data: (any scheme that can
+   *            execute or smuggle script when the href is clicked).
+   * - "src":   allow only http(s):, mailto:, /api/files paths, or a
+   *            scheme-less relative/absolute path; reject
+   *            javascript:/vbscript:/data:/file:.
+   */
+
+  /**
+   * Rename a page (change its title only) without touching or resending its
+   * content. The slug is derived from the page record, not the body, so it is
+   * left intact too.
+   */
+  async renamePage(pageId: string, title: string) {
+    await this.ensureAuthenticated();
+    await this.client.post("/pages/update", { pageId, title });
+    return { success: true, pageId, title };
+  }
+
+  /**
+   * Copy the WHOLE content of one page onto another, entirely server-side: the
+   * source's ProseMirror document is read and written verbatim onto the target
+   * via the live collab path, so the document never passes through the model.
+   *
+   * Only the target's BODY is replaced — its title and slug live on the page
+   * record (not in the content), so they are untouched. The source page is not
+   * modified at all.
+   */
+
+  async movePage(
+    pageId: string,
+    parentPageId: string | null,
+    position?: string,
+  ) {
+    await this.ensureAuthenticated();
+    // Docmost requires position >= 5 chars.
+    const validPosition = position || "a00000";
+
+    return this.client
+      .post("/pages/move", {
+        pageId,
+        parentPageId,
+        position: validPosition,
+      })
+      .then((res) => res.data);
+  }
+
+
+  async deletePage(pageId: string) {
+    await this.ensureAuthenticated();
+    return this.client
+      .post("/pages/delete", { pageId })
+      .then((res) => res.data);
+  }
+
+  // --- Comment methods (ported from upstream PR #3 by Max Nikitin) ---
+
+  /**
+   * Normalize a comment's `content` into a ProseMirror doc object before
+   * markdown conversion. createComment/updateComment send content as a
+   * JSON.stringify(...) STRING, and the server stores it as-is, so on read it
+   * comes back as a string. convertProseMirrorToMarkdown returns "" for a
+   * string, so parse it first (guarded — fall back to the raw value on any
+   * parse failure so a non-JSON legacy value is still handled gracefully).
+   */
+
+  /** Share a page publicly (idempotent) and return the public URL. */
+  async sharePage(pageId: string, searchIndexing: boolean = true) {
+    await this.ensureAuthenticated();
+    const response = await this.client.post("/shares/create", {
+      pageId,
+      includeSubPages: false,
+      searchIndexing,
+    });
+    const share = response.data?.data ?? response.data;
+    const slugId = share.page?.slugId || (await this.getPageRaw(pageId)).slugId;
+    return {
+      shareId: share.id,
+      key: share.key,
+      pageId: share.pageId,
+      publicUrl: this.shareUrl(share.key, slugId),
+      searchIndexing: share.searchIndexing,
+    };
+  }
+
+  /** List all public shares in the workspace with their URLs. */
+
+  /** Build the public share URL for a page. */
+  protected shareUrl(shareKey: string, slugId: string): string {
+    return `${this.appUrl}/share/${shareKey}/p/${slugId}`;
+  }
+
+  /** Share a page publicly (idempotent) and return the public URL. */
+
+  /** List all public shares in the workspace with their URLs. */
+  async listShares() {
+    const shares = await this.paginateAll("/shares", {});
+    return shares.map((s: any) => ({
+      shareId: s.id,
+      key: s.key,
+      pageId: s.pageId,
+      pageTitle: s.page?.title,
+      publicUrl: s.page?.slugId ? this.shareUrl(s.key, s.page.slugId) : null,
+      searchIndexing: s.searchIndexing,
+      createdAt: s.createdAt,
+    }));
+  }
+
+  /** Remove the public share of a page. */
+  async unsharePage(pageId: string) {
+    await this.ensureAuthenticated();
+    const shares = await this.listShares();
+    const share = shares.find((s: any) => s.pageId === pageId);
+    if (!share) {
+      throw new Error(`Page ${pageId} is not shared.`);
+    }
+    await this.client.post("/shares/delete", { shareId: share.shareId });
+    return { success: true, removedShareId: share.shareId, pageId };
+  }
+
+
+  /**
+   * Export a page to a single self-contained Docmost-flavoured markdown file:
+   * meta block + body (with inline comment anchors + diagrams) + comment
+   * threads. Lossless round-trip target; see importPageMarkdown for the inverse.
+   */
+  async exportPageMarkdown(pageId: string): Promise<string> {
+    await this.ensureAuthenticated();
+    const page = await this.getPageRaw(pageId);
+    const body = page.content ? convertProseMirrorToMarkdown(page.content) : "";
+    let comments: any[] = [];
+    try {
+      // Lossless export: include RESOLVED threads so the export -> import
+      // round-trip preserves every comment. This is exactly why the active-only
+      // filter is an opt-in (default false) on listComments.
+      comments = (await this.listComments(pageId, true)).items;
+    } catch (e) {
+      // A comments fetch failure must not lose the body; export with [] and let
+      // the caller see the (empty) comments block. Log under DEBUG only.
+      if (process.env.DEBUG) console.error("export: listComments failed", e);
+    }
+    const meta = {
+      version: 1,
+      pageId: page.id,
+      slugId: page.slugId,
+      title: page.title,
+      spaceId: page.spaceId,
+      parentPageId: page.parentPageId ?? null,
+    };
+    return serializeDocmostMarkdown(meta, body, comments);
+  }
+
+  /**
+   * Import a self-contained Docmost markdown file back into a page. Parses out
+   * the meta + comments metadata blocks, converts the body to ProseMirror
+   * (restoring comment marks + diagrams from their inline HTML), and replaces
+   * the page content. Comment THREAD records are NOT written to the server in
+   * this version — they are preserved in the file and the inline marks are
+   * re-applied so the highlights survive; managing comment records stays with
+   * the comment tools/UI.
+   */
+  async importPageMarkdown(pageId: string, fullMarkdown: string): Promise<any> {
+    await this.ensureAuthenticated();
+    const { meta, body, comments } = parseDocmostMarkdown(fullMarkdown);
+    // PAGE import: canonicalize footnotes (see markdownToProseMirrorCanonical).
+    const doc = await markdownToProseMirrorCanonical(body);
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+    const mutation = await replacePageContent(
+      pageUuid,
+      doc,
+      collabToken,
+      this.apiUrl,
+    );
+    // Collect distinct comment ids that actually became comment marks in the doc.
+    const collectCommentIds = (node: any, acc: Set<string>): Set<string> => {
+      if (!node || typeof node !== "object") return acc;
+      if (Array.isArray(node.marks)) {
+        for (const mk of node.marks) {
+          if (mk && mk.type === "comment" && mk.attrs?.commentId) {
+            acc.add(mk.attrs.commentId);
+          }
+        }
+      }
+      if (Array.isArray(node.content)) {
+        for (const child of node.content) collectCommentIds(child, acc);
+      }
+      return acc;
+    };
+    // Count reflects the comment marks present in the written document, so an id
+    // that only appears as inert text (e.g. inside a fenced code block) is not
+    // counted because it never becomes a comment mark.
+    const anchoredIds = collectCommentIds(doc, new Set<string>());
+    const result: any = {
+      success: true,
+      pageId,
+      anchoredCommentCount: anchoredIds.size,
+      commentsInFile: Array.isArray(comments) ? comments.length : 0,
+      verify: mutation.verify,
+    };
+    // Warn (non-fatal) if the file was exported from a DIFFERENT page.
+    if (meta?.pageId && meta.pageId !== pageId) {
+      result.warning = `File was exported from page ${meta.pageId} but is being imported into ${pageId}.`;
+    }
+    // Non-fatal footnote diagnostics (#166), analyzed on the BODY (the part after
+    // the docmost:meta / docmost:comments blocks) — so a `[^x]`-like token inside
+    // those JSON blocks never produces a false warning, while real markers in the
+    // body do. `body` comes from parseDocmostMarkdown(fullMarkdown) above.
+    Object.assign(result, footnoteWarningsField(body));
+    return result;
+  }
+
+  /**
+   * Rename a page (change its title only) without touching or resending its
+   * content. The slug is derived from the page record, not the body, so it is
+   * left intact too.
+   */
+
+  /**
+   * Copy the WHOLE content of one page onto another, entirely server-side: the
+   * source's ProseMirror document is read and written verbatim onto the target
+   * via the live collab path, so the document never passes through the model.
+   *
+   * Only the target's BODY is replaced — its title and slug live on the page
+   * record (not in the content), so they are untouched. The source page is not
+   * modified at all.
+   */
+  async copyPageContent(sourcePageId: string, targetPageId: string) {
+    await this.ensureAuthenticated();
+
+    // A self-copy would be a no-op overwrite; reject it explicitly so a caller
+    // mistake surfaces as a clear error rather than a silent round-trip.
+    if (sourcePageId === targetPageId) {
+      throw new Error(
+        "copyPageContent: sourcePageId and targetPageId are the same page (no-op copy)",
+      );
+    }
+
+    const source = await this.getPageRaw(sourcePageId);
+    const content = source?.content;
+    if (
+      !content ||
+      typeof content !== "object" ||
+      content.type !== "doc" ||
+      !Array.isArray(content.content)
+    ) {
+      throw new Error(
+        `copyPageContent: source page ${sourcePageId} has no usable ProseMirror content to copy`,
+      );
+    }
+
+    // Defense-in-depth: run the same URL-scheme sanitizer the JSON write path
+    // uses, so copying never lands a javascript:/data: href/src on the target
+    // (parity with updatePageJson; harmless for already-stored source content).
+    this.validateDocUrls(content);
+
+    // Defense-in-depth (#228): this is a FULL-document write, so canonicalize
+    // footnotes before copying — a no-op on already-canonical source content, but
+    // it guarantees a copy can never propagate a non-canonical footnote topology
+    // to the target (parity with the other full-doc write paths).
+    // #419: normalize + merge glyph-forked definitions before canonicalizing.
+    const canonical = canonicalizeFootnotes(normalizeAndMergeFootnotes(content));
+
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the TARGET collab doc by its canonical UUID, never the slugId (#260).
+    const targetUuid = await this.resolvePageId(targetPageId);
+    const mutation = await this.replacePage(
+      targetUuid,
+      canonical,
+      collabToken,
+      this.apiUrl,
+    );
+
+    return {
+      success: true,
+      sourcePageId,
+      targetPageId,
+      copiedNodes: canonical.content.length,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Surgical text edits: find/replace inside text nodes of the live
+   * document. Preserves all block ids, marks, callouts and tables.
+   */
+
+  // --- Page history / diff / transform ---
+
+  /**
+   * List the saved versions (history snapshots) of a page, newest first.
+   * Docmost auto-snapshots on every save. Returns one cursor-paginated page of
+   * results: `{ items, nextCursor }`. The history record's id field is `id`.
+   */
+  async listPageHistory(pageId: string, cursor?: string) {
+    await this.ensureAuthenticated();
+    const payload: Record<string, any> = { pageId };
+    if (cursor) payload.cursor = cursor;
+    const response = await this.client.post("/pages/history", payload);
+    const data = response.data?.data ?? response.data;
+    return {
+      items: data?.items ?? [],
+      nextCursor: data?.meta?.nextCursor ?? null,
+    };
+  }
+
+  /**
+   * Fetch a single page-history version including its lossless ProseMirror
+   * `content`. The version also carries pageId/title/createdAt.
+   */
+  async getPageHistory(historyId: string) {
+    await this.ensureAuthenticated();
+    const response = await this.client.post("/pages/history/info", {
+      historyId,
+    });
+    return response.data?.data ?? response.data;
+  }
+
+  /**
+   * "Restore" a version: Docmost has NO restore endpoint, so we take the
+   * version's `content` and write it as the page's current content via the live
+   * collab path (which itself creates a new history snapshot). Returns the
+   * affected pageId and the source historyId.
+   */
+  async restorePageVersion(historyId: string) {
+    await this.ensureAuthenticated();
+    const version = await this.getPageHistory(historyId);
+    if (
+      !version ||
+      !version.pageId ||
+      !version.content ||
+      typeof version.content !== "object"
+    ) {
+      throw new Error(
+        `restorePageVersion: history ${historyId} has no usable content`,
+      );
+    }
+    // Defense-in-depth: sanitize URLs in the restored content (parity with the
+    // JSON write path) before writing it back.
+    this.validateDocUrls(version.content);
+    const collabToken = await this.getCollabTokenWithReauth();
+    // version.pageId is the page entity id (already a UUID); resolvePageId
+    // short-circuits a UUID with no round-trip, so this is defensive only (#260).
+    const pageUuid = await this.resolvePageId(version.pageId);
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      () => version.content,
+    );
+    return {
+      pageId: version.pageId,
+      restoredFrom: historyId,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Diff two versions of a page and return a Docmost-equivalent change set.
+   * `from`/`to` each resolve to a ProseMirror doc:
+   *   - null / undefined / "current" -> the page's CURRENT content;
+   *   - any other string             -> that historyId's content.
+   * Returns the diff plus the resolved version metadata for each side.
+   */
+  async diffPageVersions(pageId: string, from?: string, to?: string) {
+    await this.ensureAuthenticated();
+
+    const isCurrent = (v?: string) => v == null || v === "" || v === "current";
+
+    const resolveSide = async (
+      v?: string,
+    ): Promise<{ doc: any; meta: any }> => {
+      if (isCurrent(v)) {
+        const raw = await this.getPageRaw(pageId);
+        return {
+          doc: raw.content || { type: "doc", content: [] },
+          meta: {
+            kind: "current",
+            pageId,
+            title: raw.title,
+            updatedAt: raw.updatedAt,
+          },
+        };
+      }
+      const version = await this.getPageHistory(v as string);
+      return {
+        doc: version.content || { type: "doc", content: [] },
+        meta: {
+          kind: "history",
+          historyId: version.id,
+          pageId: version.pageId,
+          title: version.title,
+          createdAt: version.createdAt,
+        },
+      };
+    };
+
+    const fromSide = await resolveSide(from);
+    const toSide = await resolveSide(to);
+    const diff = diffDocs(fromSide.doc, toSide.doc);
+    return { from: fromSide.meta, to: toSide.meta, diff };
+  }
+
+  /**
+   * Edit a page by running an arbitrary user-supplied JS transform against the
+   * live document, with a diff preview + page-history safety net.
+   *
+   * The transform string is evaluated as `(doc, ctx) => doc` inside a node:vm
+   * sandbox: it gets ONLY `{ doc, ctx, structuredClone, console }` as globals,
+   * a 5s timeout, and NO access to require/process/fs/network. It must return a
+   * `{ type: "doc" }` node, which is validated structurally before any write.
+   *
+   * `ctx` exposes:
+   *   - comments: the page's comments (fetched before the live read);
+   *   - log: an array the transform can push diagnostics to (via console.log);
+   *   - consume(id): mark a comment id as consumed (for deleteComments);
+   *   - helpers: the transforms.ts primitives + commentsToFootnotes.
+   *
+   * Footnote convention used by the helpers: footnote markers are plain "[N]"
+   * text in the body, and the notes are an orderedList under a heading whose
+   * text is "Примечания переводчика".
+   *
+   * dryRun (default true): read the page's current content, run the transform,
+   * and return `{ pushed:false, diff, log }` WITHOUT opening the collab socket.
+   * Otherwise the transform runs atomically inside mutatePageContent, optionally
+   * deletes consumed comments, and returns the new historyId + diff + log.
+   */
+  }
+  return PagesMixin;
+}

packages/mcp/src/client/read.ts

@@ -0,0 +1,713 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import axios, { AxiosInstance } from "axios";
+import {
+  filterWorkspace,
+  filterSpace,
+  filterPage,
+  filterComment,
+  filterSearchResult,
+} from "../lib/filters.js";
+import {
+  convertProseMirrorToMarkdown,
+  type ConvertProseMirrorToMarkdownOptions,
+} from "../lib/markdown-converter.js";
+import {
+  GetPageConversionCache,
+  hashConvertOptions,
+} from "./getpage-cache.js";
+import {
+  collectInternalFileNodes,
+  normalizeFileUrl,
+  resolveInternalFilePath,
+} from "../lib/internal-file-urls.js";
+import { buildPageTree } from "../lib/tree.js";
+import {
+  replaceNodeById,
+  replaceNodeByIdWithMany,
+  reassignCollidingBlockIds,
+  deleteNodeById,
+  assertUnambiguousMatch,
+  insertNodeRelative,
+  insertNodesRelative,
+  blockPlainText,
+  buildOutline,
+  getNodeByRef,
+  readTable,
+  insertTableRow,
+  deleteTableRow,
+  updateTableCell,
+  findInvalidNode,
+} from "@docmost/prosemirror-markdown";
+import {
+  importMarkdownFragment,
+  canBeDocChild,
+  findUnrepresentableTableAttrs,
+} from "../lib/markdown-fragment.js";
+import { searchInDoc, SearchOptions } from "../lib/page-search.js";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+
+// Public method surface of ReadMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements IReadMixin` fails to compile on drift.
+export interface IReadMixin {
+  getWorkspace(): any;
+  getSpaces(): any;
+  listPages(spaceId?: string, limit?: number, tree?: boolean): any;
+  getTree(spaceId: string, rootPageId?: string, maxDepth?: number): any;
+  getPageContext(pageId: string): any;
+  listSidebarPages(spaceId: string, pageId?: string): any;
+  getPage(pageId: string): any;
+  getPageJson(pageId: string): any;
+  getOutline(pageId: string): any;
+  getNode(pageId: string, nodeId: string, format?: "markdown" | "json"): any;
+  searchInPage(pageId: string, query: string, opts?: SearchOptions): any;
+  getTable(pageId: string, tableRef: string): any;
+  search(query: string, spaceId?: string, limit?: number, opts?: { parentPageId?: string; titleOnly?: boolean }): any;
+}
+
+export function ReadMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & IReadMixin> & TBase {
+  abstract class ReadMixin extends Base implements IReadMixin {
+  async getWorkspace() {
+    await this.ensureAuthenticated();
+    const response = await this.client.post("/workspace/info", {});
+    return {
+      data: filterWorkspace(response.data?.data ?? response.data),
+      success: response.data.success,
+    };
+  }
+
+
+  async getSpaces() {
+    const spaces = await this.paginateAll("/spaces", {});
+    return spaces.map((space) => filterSpace(space));
+  }
+
+  /**
+   * List pages in one of two modes.
+   *
+   * Default (`tree` false): most recent pages by updatedAt (descending),
+   * bounded. Fetching the whole space can exceed MCP response/time limits on
+   * large instances, so a single bounded page of results is returned (default
+   * 50, max 100) via the `/pages/recent` feed.
+   *
+   * Tree (`tree` true): DEPRECATED — prefer `getTree`, which shares this exact
+   * code path (a single `/pages/tree` request via `enumerateSpacePages` +
+   * `buildPageTree`) but returns the compact `{pageId, title, children?,
+   * hasChildren?}` shape and supports `rootPageId`/`maxDepth`. This tree mode is
+   * kept for backward compatibility; it REQUIRES `spaceId` (a page tree is
+   * scoped to one space) and IGNORES `limit` — the whole hierarchy is returned.
+   * It fetches the tree via `enumerateSpacePages`, which on the fork server
+   * resolves to a single `/pages/tree` request returning the whole
+   * permission-filtered flat page set (soft-deleted pages excluded
+   * server-side); the cursor-BFS in `enumerateSpacePages` is only a fallback for
+   * stock upstream servers that lack `/pages/tree`.
+   */
+  async listPages(spaceId?: string, limit: number = 50, tree: boolean = false) {
+    await this.ensureAuthenticated();
+
+    if (tree) {
+      if (!spaceId) {
+        throw new Error(
+          "listPages: tree mode requires a spaceId (a page tree is scoped to one space). Pass spaceId, or omit tree to get the recent-pages list.",
+        );
+      }
+      const { pages } = await this.enumerateSpacePages(spaceId);
+      return buildPageTree(pages);
+    }
+
+    const clampedLimit = Math.max(1, Math.min(100, limit));
+    const payload: Record<string, any> = { limit: clampedLimit, page: 1 };
+    if (spaceId) payload.spaceId = spaceId;
+    const response = await this.client.post("/pages/recent", payload);
+    const data = response.data;
+    const items = data.data?.items || data.items || [];
+    return items.map((page: any) => filterPage(page));
+  }
+
+  /**
+   * Fetch a space's page hierarchy (or one subtree) as a nested tree in a SINGLE
+   * request — the #443 `getTree` tool. Shares its whole code path with
+   * `listPages(tree:true)`: `enumerateSpacePages` issues one `POST /pages/tree`
+   * (with the cursor-BFS only as a fallback for stock upstream servers that lack
+   * the endpoint), then `buildPageTree` nests the flat, permission-filtered,
+   * position-ordered list. No second tree fetch, no per-node BFS.
+   *
+   *  - `rootPageId` — restrict to that page's subtree; the server seeds the CTE
+   *    with the page itself, so the result is exactly ONE root (the page and its
+   *    descendants). Omit it for the whole space.
+   *  - `maxDepth` — trim the response to that many levels (roots = depth 1) to
+   *    save tokens; the server still returns everything in one request, the cut
+   *    is applied in `buildPageTree` AFTER the full tree is built. A node whose
+   *    children were cut carries `hasChildren: true` (source of truth = the flat
+   *    item's server `hasChildren`) so the caller can descend with a follow-up
+   *    `getTree(spaceId, rootPageId=that node)` call.
+   *
+   * Output nodes are `{pageId, title, children?, hasChildren?}` — only the UUID
+   * `pageId` is exposed (never `slugId`/`icon`/`position`). Requires `spaceId`
+   * (a page tree is scoped to one space).
+   */
+  async getTree(spaceId: string, rootPageId?: string, maxDepth?: number) {
+    await this.ensureAuthenticated();
+    if (!spaceId) {
+      throw new Error(
+        "getTree: spaceId is required (a page tree is scoped to one space).",
+      );
+    }
+    const { pages } = await this.enumerateSpacePages(spaceId, rootPageId);
+    return buildPageTree(pages, { shape: "getTree", maxDepth });
+  }
+
+  /**
+   * "Where am I / what's around" for a single page — the #443 `getPageContext`
+   * tool. Metadata only (no page content), using exactly TWO server requests:
+   *
+   *   1. `POST /pages/breadcrumbs` — a recursive CTE that walks UP from the page.
+   *      The server returns the chain root->page order (it `.reverse()`s the
+   *      child-first walk before responding), INCLUDING the page itself as the
+   *      LAST element. So the last element is the page and everything before it
+   *      is the ancestor chain root->parent. This carries the page's own title
+   *      and spaceId, so no extra page-info fetch is needed for a UUID input.
+   *   2. `listSidebarPages(spaceId, pageId)` — the page's DIRECT children,
+   *      cursor-paginated (a page with >20 children returns ALL of them, no
+   *      dupes) and in sidebar `position` order, each carrying `hasChildren`.
+   *
+   * The input may be a slugId (agents copy them from URLs); it is run through
+   * `resolvePageId` first, exactly like the other page tools. A UUID input adds
+   * no request there (short-circuit), keeping the total at two; a slugId input
+   * adds one unavoidable resolve round-trip.
+   *
+   * INVARIANT: only the UUID `pageId` is exposed anywhere — server `id` is
+   * mapped to `pageId` and `slugId` is never leaked. A nonexistent/inaccessible
+   * pageId makes the server 404/403, which propagates as a clear tool error
+   * (never a hollow empty object).
+   */
+  async getPageContext(pageId: string) {
+    await this.ensureAuthenticated();
+
+    // Resolve a possibly-slugId input to the canonical UUID (no round-trip for a
+    // UUID). Errors here (bad/inaccessible id) propagate as a clear tool error.
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Request 1: the ancestor chain, root->page, page included as the LAST item.
+    const response = await this.client.post("/pages/breadcrumbs", {
+      pageId: pageUuid,
+    });
+    const chain: any[] = (response.data?.data ?? response.data) ?? [];
+    if (!Array.isArray(chain) || chain.length === 0) {
+      // The endpoint always includes the page itself, so an empty chain means
+      // the page is gone/inaccessible — surface a clear error, not {}.
+      throw new Error(`getPageContext: page "${pageId}" not found or inaccessible`);
+    }
+
+    // Split: the last element is the page, the rest (root->parent) are the
+    // breadcrumbs. A root page has no ancestors -> breadcrumbs is [].
+    const self = chain[chain.length - 1];
+    const ancestors = chain.slice(0, -1);
+
+    const page = {
+      pageId: self.id,
+      title: self.title,
+      spaceId: self.spaceId,
+    };
+    const breadcrumbs = ancestors.map((n: any) => ({
+      pageId: n.id,
+      title: n.title,
+    }));
+
+    // Request 2: direct children in sidebar order, each with hasChildren.
+    const childItems = await this.listSidebarPages(self.spaceId, pageUuid);
+    const children = childItems.map((c: any) => ({
+      pageId: c.id,
+      title: c.title,
+      hasChildren: Boolean(c.hasChildren),
+    }));
+
+    return { page, breadcrumbs, children };
+  }
+
+  /**
+   * List sidebar pages for a space. With no pageId the request returns the
+   * space ROOT pages; with a pageId it returns the direct CHILDREN of that
+   * page. pageId is therefore optional and is only included in the POST body
+   * when provided (an empty/undefined pageId would otherwise change the
+   * semantics on the server).
+   */
+  async listSidebarPages(spaceId: string, pageId?: string) {
+    await this.ensureAuthenticated();
+
+    // Paginate via the server-issued cursor. The server switched from OFFSET
+    // (`page`) to CURSOR (`cursor`/`nextCursor`) pagination, and the global
+    // ValidationPipe(whitelist:true) SILENTLY STRIPS the obsolete `page` field
+    // — so the old offset loop got the SAME first page every time (with
+    // hasNextPage stuck true) and dropped every child beyond the first page.
+    const MAX_PAGES = 50;
+    let cursor: string | undefined;
+    let allItems: any[] = [];
+    let truncated = false;
+
+    for (let i = 0; i < MAX_PAGES; i++) {
+      // limit: 100 is the server-side Max; cuts request count 5x vs the default 20.
+      const payload: Record<string, any> = { spaceId, limit: 100 };
+      // Only send pageId when scoping to a page's children; omit it for roots.
+      if (pageId) payload.pageId = pageId;
+      if (cursor) payload.cursor = cursor;
+
+      const data = (await this.client.post("/pages/sidebar-pages", payload)).data
+        ?.data;
+      allItems = allItems.concat(data?.items ?? []);
+
+      // Advance strictly via the server-issued cursor; a missing/repeated cursor
+      // means the protocol drifted again — stop instead of looping on page one.
+      const next = data?.meta?.hasNextPage ? data?.meta?.nextCursor : null;
+      if (!next || next === cursor) break;
+      cursor = next;
+
+      // Reaching the ceiling with more pages still available means the child
+      // list is truncated (mirrors paginateAll).
+      if (i === MAX_PAGES - 1) truncated = true;
+    }
+
+    // Warn on real truncation (ceiling hit while the server still had pages) so
+    // the caller is not silently handed an incomplete child list.
+    if (truncated) {
+      console.warn(
+        `listSidebarPages: children of "${pageId ?? spaceId}" truncated at the ${MAX_PAGES}-page cap; more pages exist on the server`,
+      );
+    }
+
+    return allItems;
+  }
+
+  /**
+   * Enumerate EVERY page in a space (or in a subtree, when rootPageId is given).
+   *
+   * Primary path (fork server): a SINGLE `POST /pages/tree` returns the whole
+   * space (or a subtree) as a flat, permission-filtered list in one request, in
+   * the exact node shape buildPageTree consumes. This replaces the old
+   * per-node BFS, which issued N sidebar requests and — after the server moved
+   * to cursor pagination — silently lost every child past the first sidebar
+   * page (the obsolete `page` param was stripped by ValidationPipe).
+   *
+   * The subtree variant (rootPageId given) INCLUDES the root node itself
+   * (getPageAndDescendants seeds with id = rootPageId), unlike the old BFS
+   * which started from the root's children.
+   *
+   * Fallback path (stdio mode may target STOCK upstream Docmost, which lacks
+   * `/pages/tree`): on a 404/405 it falls back to the cursor-based BFS below,
+   * walking direct children via the fixed cursor listSidebarPages. Safeguards:
+   * a `visited` Set of page ids prevents re-processing a node (cycles /
+   * duplicate references), and a hard node cap bounds pathological trees so the
+   * walk always terminates.
+   *
+   * Returns `{ pages, truncated }`. `truncated` is true ONLY when the fallback
+   * BFS stopped at its MAX_NODES cap — the primary /pages/tree path is uncapped
+   * and always returns the complete set, so it never reports truncation.
+   */
+  protected async enumerateSpacePages(
+    spaceId: string,
+    rootPageId?: string,
+  ): Promise<{ pages: any[]; truncated: boolean }> {
+    await this.ensureAuthenticated();
+
+    // Single request replaces the whole BFS: /pages/tree returns the full
+    // permission-filtered flat page set of a space (or a subtree) at once. This
+    // path is uncapped, so it is never truncated.
+    const payload = rootPageId ? { pageId: rootPageId } : { spaceId };
+    try {
+      const response = await this.client.post("/pages/tree", payload);
+      const pages = (response.data?.data ?? response.data)?.items ?? [];
+      return { pages, truncated: false };
+    } catch (e: any) {
+      // Only fall back when the endpoint is absent (stock upstream Docmost);
+      // any other error is a genuine failure and must propagate.
+      if (
+        !axios.isAxiosError(e) ||
+        (e.response?.status !== 404 && e.response?.status !== 405)
+      ) {
+        throw e;
+      }
+    }
+
+    // Fallback: cursor-based breadth-first walk via listSidebarPages.
+    const MAX_NODES = 10000;
+    const result: any[] = [];
+    const visited = new Set<string>();
+
+    // Seed with the root node itself when scoping to a subtree, so its own
+    // comments aren't dropped: the primary /pages/tree seeds
+    // getPageAndDescendants with id = rootPageId (root included), but
+    // listSidebarPages(spaceId, rootPageId) returns only the root's CHILDREN.
+    // The `visited` set below prevents a double-add if the root also appears
+    // among the children. getPageRaw returns a page whose id/title/spaceId are
+    // exactly what buildPageTree and checkNewComments consume.
+    if (rootPageId) {
+      try {
+        const root = await this.getPageRaw(rootPageId);
+        if (root?.id) {
+          result.push(root);
+          visited.add(root.id);
+        }
+      } catch {
+        // Non-fatal: if the root can't be read, fall through to children-only.
+      }
+    }
+
+    // Seed the queue with the starting level (subtree children or roots).
+    const queue: any[] = await this.listSidebarPages(spaceId, rootPageId);
+
+    while (queue.length > 0 && result.length < MAX_NODES) {
+      const node = queue.shift();
+      if (!node || typeof node !== "object" || !node.id) continue;
+
+      // Skip already-seen ids to guard against cycles / duplicate references.
+      if (visited.has(node.id)) continue;
+      visited.add(node.id);
+
+      result.push(node);
+
+      if (node.hasChildren) {
+        try {
+          const children = await this.listSidebarPages(spaceId, node.id);
+          for (const child of children) queue.push(child);
+        } catch (e: any) {
+          // A failure fetching one node's children must not abort the whole
+          // walk: skip this branch and keep enumerating the rest.
+        }
+      }
+    }
+
+    // Truncated only when the cap was hit with the queue still non-empty (real
+    // truncation, not a natural end at exactly MAX_NODES).
+    return {
+      pages: result,
+      truncated: result.length >= MAX_NODES && queue.length > 0,
+    };
+  }
+
+  /** Raw page info including the ProseMirror JSON content and slugId. */
+
+  /**
+   * Overridable seam over convertProseMirrorToMarkdown (issue #479). Production
+   * just delegates; it exists as a method so a unit test can spy on it and
+   * assert the conversion is genuinely SKIPPED on a getPage cache HIT (the whole
+   * point of the cache) — an ESM named import cannot be intercepted otherwise.
+   */
+  protected convertPageMarkdown(
+    content: any,
+    options: ConvertProseMirrorToMarkdownOptions,
+  ): string {
+    return convertProseMirrorToMarkdown(content, options);
+  }
+
+  async getPage(pageId: string) {
+    await this.ensureAuthenticated();
+    const resultData = await this.getPageRaw(pageId);
+
+    // Agent read: hide resolved-comment anchors so the agent sees only active
+    // discussions. Active anchors are kept. (The lossless exportPageMarkdown
+    // round-trip deliberately does NOT pass this flag — resolved anchors there
+    // must be preserved.)
+    //
+    // Content-addressed conversion cache (issue #479): the PM->Markdown walk is
+    // the dominant cost of this hot read op. Key on the page's canonical UUID +
+    // updatedAt (both from THIS /pages/info response, so mutually consistent) +
+    // a hash of the conversion options. A hit returns the cached markdown and
+    // skips the walk; a miss converts and stores. The cached value is the
+    // conversion output BEFORE the {{SUBPAGES}} substitution below, which uses
+    // live subpage data and stays outside the cache — so the final result is
+    // byte-identical to the uncached path.
+    const convertOptions = { dropResolvedCommentAnchors: true };
+    let content = "";
+    if (resultData.content) {
+      // Only cache when we have a stable identity+version for the key. Both come
+      // from the same response; if either is missing (unexpected server shape),
+      // fall back to converting uncached rather than keying on a partial tuple.
+      const cacheable =
+        typeof resultData.id === "string" &&
+        typeof resultData.updatedAt === "string";
+      const cacheKey = cacheable
+        ? GetPageConversionCache.key(
+            resultData.id,
+            resultData.updatedAt,
+            hashConvertOptions(convertOptions),
+          )
+        : null;
+
+      const cached = cacheKey ? this.getPageCache.get(cacheKey) : undefined;
+      if (cached !== undefined) {
+        content = cached;
+        this.onMetricFn?.("mcp_getpage_cache_hits_total", 1);
+      } else {
+        // Goes through the convertPageMarkdown seam (not the raw import) so a
+        // test can assert the conversion is SKIPPED on a hit (issue #479 F2).
+        content = this.convertPageMarkdown(resultData.content, convertOptions);
+        if (cacheKey) this.getPageCache.set(cacheKey, content);
+        // A non-cacheable page (missing id/updatedAt) is still a genuine
+        // conversion, so it counts as a miss for an honest hit-rate.
+        this.onMetricFn?.("mcp_getpage_cache_misses_total", 1);
+      }
+    }
+
+    // Always fetch subpages to provide context to the agent.
+    //
+    // NOT parallelizable with the page fetch (issue #479 asked to check): the
+    // sidebar-pages endpoint REQUIRES spaceId in its POST body, and spaceId is
+    // only known FROM this page fetch's response (resolvePageId yields the UUID
+    // but never the spaceId). So `Promise.all([pageFetch, subpagesFetch])` would
+    // have to invent a spaceId it does not have — the two calls are inherently
+    // sequential. Correctness wins; the conversion cache above is the real speedup.
+    let subpages: any[] = [];
+    try {
+      // `pageId` may be a slugId, but the sidebar-pages endpoint requires the
+      // UUID; `resultData.id` holds the resolved UUID returned by getPageRaw.
+      subpages = await this.listSidebarPages(resultData.spaceId, resultData.id);
+    } catch (e: any) {
+      console.warn("Failed to fetch subpages:", e);
+    }
+
+    // Resolve subpages if the placeholder exists
+    if (content && content.includes("{{SUBPAGES}}")) {
+      if (subpages && subpages.length > 0) {
+        const list = subpages
+          .map((p: any) => `- [${p.title}](page:${p.id})`)
+          .join("\n");
+        content = content.replace("{{SUBPAGES}}", `### Subpages\n${list}`);
+      } else {
+        content = content.replace("{{SUBPAGES}}", "");
+      }
+    }
+
+    return {
+      data: filterPage(resultData, content, subpages),
+      success: true,
+    };
+  }
+
+  /** Page info + raw ProseMirror JSON content (lossless representation). */
+  async getPageJson(pageId: string) {
+    const data = await this.getPageRaw(pageId);
+    return {
+      id: data.id,
+      slugId: data.slugId,
+      title: data.title,
+      parentPageId: data.parentPageId,
+      spaceId: data.spaceId,
+      updatedAt: data.updatedAt,
+      content: data.content || { type: "doc", content: [] },
+    };
+  }
+
+  /**
+   * Fetch an INTERNAL Docmost file (authed loopback) for sandbox mirroring.
+   * `src` is normalized to `/api/files/<id>/<file>`; `this.client.baseURL`
+   * already ends in `/api`, so we strip the leading `/api` and request the
+   * relative path with the client's Authorization header. Returns the raw bytes
+   * and the response Content-Type (mime), defaulting to octet-stream.
+   *
+   * The fetch is size-bounded (hard 64 MiB ceiling) purely to protect memory;
+   * the authoritative per-blob cap is enforced by the sandbox `put`. The path is
+   * resolved via resolveInternalFilePath, which REJECTS (throws) any traversal
+   * or percent-encoded src that would let an attacker-controlled `attrs.src`
+   * escape `/api/files/` and reach another internal endpoint (SSRF). That throw
+   * happens before this.client.get, so a malicious src is counted as a failed
+   * mirror — it never reaches the network.
+   */
+
+  /**
+   * Compact outline of a page's top-level blocks (no full document body).
+   * Cheap way to locate sections/tables and grab block ids before drilling in
+   * with getNode / patchNode / insertNode.
+   */
+  async getOutline(pageId: string) {
+    await this.ensureAuthenticated();
+    const data = await this.getPageRaw(pageId);
+    return {
+      pageId,
+      slugId: data.slugId,
+      title: data.title,
+      outline: buildOutline(data.content ?? { type: "doc", content: [] }),
+    };
+  }
+
+  /**
+   * Fetch a single block for editing by reference: a block id (headings/
+   * paragraphs/callouts/images), or `#<index>` to select a top-level block by its
+   * outline index (the only way to reach tables/rows/cells, which carry no id).
+   *
+   * `format` (#413):
+   *  - `"markdown"` (DEFAULT): serialize the block via the canonical converter
+   *    (`{type:"doc",content:[node]}` -> `convertProseMirrorToMarkdown`) — a read
+   *    "for editing": pair it with `patchNode({markdown})` to rewrite the block.
+   *    Comment anchors (`<span data-comment-id>`, INCLUDING resolved ones) are
+   *    NOT stripped here (unlike getPage): losing them on write-back would
+   *    orphan the thread. Returns `{ ..., format:"markdown", markdown }`.
+   *  - `"json"`: return the raw ProseMirror subtree as-is (lossless; the previous
+   *    default). Returns `{ ..., format:"json", node }`.
+   *
+   * AUTO fallback: a type that cannot be a document top-level child
+   * (tableRow/tableCell/tableHeader, addressed by `#<index>`) is NOT expressible
+   * as a standalone markdown document, so a `"markdown"` request for such a node
+   * transparently falls back to JSON with an explicit `format:"json"` field. The
+   * check derives from the schema's `doc` contentMatch, so it tracks the schema.
+   */
+  async getNode(
+    pageId: string,
+    nodeId: string,
+    format: "markdown" | "json" = "markdown",
+  ) {
+    await this.ensureAuthenticated();
+    const data = await this.getPageRaw(pageId);
+    const hit = getNodeByRef(
+      data.content ?? { type: "doc", content: [] },
+      nodeId,
+    );
+    if (!hit) {
+      throw new Error(
+        `getNode: no node found for "${nodeId}" on page ${pageId} (use a block id from getOutline, or "#<index>" for a top-level block such as a table)`,
+      );
+    }
+
+    // JSON requested (or a non-top-level type that markdown cannot represent as a
+    // standalone document): return the subtree verbatim.
+    if (format === "json" || !canBeDocChild(hit.type)) {
+      return {
+        pageId,
+        ref: nodeId,
+        path: hit.path,
+        type: hit.type,
+        format: "json" as const,
+        node: hit.node,
+      };
+    }
+
+    // Markdown: wrap the node as a one-block doc and run the canonical converter.
+    // Comment anchors are DELIBERATELY preserved (converter default) so a
+    // getNode(markdown) -> edit -> patchNode(markdown) round trip does not orphan
+    // a comment thread; this differs from getPage, which strips them.
+    const markdown = convertProseMirrorToMarkdown({
+      type: "doc",
+      content: [hit.node],
+    });
+    return {
+      pageId,
+      ref: nodeId,
+      path: hit.path,
+      type: hit.type,
+      format: "markdown" as const,
+      markdown,
+    };
+  }
+
+  /**
+   * Find every occurrence of `query` on a page IN MEMORY, over the plain text of
+   * each text container (reusing the same `getPageRaw` fetch as the other read
+   * tools) — no server search endpoint, no whole-document round-trip through the
+   * model. Returns `{ total, truncated, matches }`; each match carries a ref for
+   * getNode/patchNode (the `#<index>` form resolves with getNode but NOT
+   * patchNode — see SearchMatch.nodeId), plus the top-level block index and a
+   * short context window used to build a unique text `selection` for
+   * createComment (createComment has no nodeId param). The pure engine
+   * (`searchInDoc`) owns the traversal, glue, the RE2 ReDoS-safe regex engine
+   * and the empty-query / invalid-or-unsupported-regex errors.
+   */
+  async searchInPage(pageId: string, query: string, opts: SearchOptions = {}) {
+    await this.ensureAuthenticated();
+    const data = await this.getPageRaw(pageId);
+    const result = searchInDoc(
+      data.content ?? { type: "doc", content: [] },
+      query,
+      opts,
+    );
+    return { pageId, query, ...result };
+  }
+
+  /**
+   * Read a table as a matrix. `tableRef` is `#<index>` (from getOutline) or a
+   * block id of any node inside the table. Returns the cell texts plus a
+   * parallel cellIds matrix (each cell's first paragraph id, or null) so a
+   * caller can patchNode a cell for rich-formatted edits. Throws when no table
+   * resolves for the reference.
+   */
+  async getTable(pageId: string, tableRef: string) {
+    await this.ensureAuthenticated();
+    const data = await this.getPageRaw(pageId);
+    const t = readTable(data.content ?? { type: "doc", content: [] }, tableRef);
+    if (!t) {
+      throw new Error(
+        `tableGet: no table found for "${tableRef}" on page ${pageId} (use "#<index>" from getOutline, or a block id inside the table)`,
+      );
+    }
+    return {
+      pageId,
+      table: tableRef,
+      rows: t.rows,
+      cols: t.cols,
+      path: t.path,
+      cells: t.cells,
+      cellIds: t.cellIds,
+    };
+  }
+
+  /**
+   * Insert a row of plain-text cells into a table on the LIVE collab document.
+   * `tableRef` is `#<index>` or a block id inside the target table. `cells` is
+   * padded to the table's column count (more cells than columns throws); `index`
+   * is a 0-based insert position (omit/out-of-range to append). Throws when no
+   * table resolves for the reference.
+   */
+
+  async search(
+    query: string,
+    spaceId?: string,
+    limit?: number,
+    opts: { parentPageId?: string; titleOnly?: boolean } = {},
+  ) {
+    await this.ensureAuthenticated();
+    // Opt into the #443 agent-lookup mode: `substring: true` turns on the hybrid
+    // substring + FTS branch that returns path + snippet + score. A stock
+    // upstream server strips these unknown DTO fields (whitelist:true) and
+    // silently degrades to plain FTS — see the tool-registration comment.
+    const payload: Record<string, any> = {
+      query,
+      spaceId,
+      substring: true,
+    };
+    if (opts.parentPageId) payload.parentPageId = opts.parentPageId;
+    if (opts.titleOnly) payload.titleOnly = true;
+    // Clamp an optional caller-supplied limit into the lookup range (1..50)
+    // before forwarding; omit it when not provided so the server default applies.
+    if (limit !== undefined) {
+      payload.limit = Math.max(1, Math.min(50, limit));
+    }
+    const response = await this.client.post("/search", payload);
+
+    // Normalize both response shapes: bare array and paginated { items: [...] }
+    const data = response.data?.data;
+    const items = Array.isArray(data) ? data : data?.items || [];
+    const filteredItems = items.map((item: any) => filterSearchResult(item));
+
+    return {
+      items: filteredItems,
+      success: response.data?.success || false,
+    };
+  }
+
+  }
+  return ReadMixin;
+}

packages/mcp/src/client/stash.ts

@@ -0,0 +1,225 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import {
+  collectInternalFileNodes,
+  normalizeFileUrl,
+  resolveInternalFilePath,
+} from "../lib/internal-file-urls.js";
+
+// Public method surface of StashMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements IStashMixin` fails to compile on drift.
+export interface IStashMixin {
+  stashPage(pageId: string): Promise<{ uri: string; sha256: string; size: number; images: { mirrored: number; failed: number }; }>;
+}
+
+export function StashMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & IStashMixin> & TBase {
+  abstract class StashMixin extends Base implements IStashMixin {
+  /**
+   * Fetch an INTERNAL Docmost file (authed loopback) for sandbox mirroring.
+   * `src` is normalized to `/api/files/<id>/<file>`; `this.client.baseURL`
+   * already ends in `/api`, so we strip the leading `/api` and request the
+   * relative path with the client's Authorization header. Returns the raw bytes
+   * and the response Content-Type (mime), defaulting to octet-stream.
+   *
+   * The fetch is size-bounded (hard 64 MiB ceiling) purely to protect memory;
+   * the authoritative per-blob cap is enforced by the sandbox `put`. The path is
+   * resolved via resolveInternalFilePath, which REJECTS (throws) any traversal
+   * or percent-encoded src that would let an attacker-controlled `attrs.src`
+   * escape `/api/files/` and reach another internal endpoint (SSRF). That throw
+   * happens before this.client.get, so a malicious src is counted as a failed
+   * mirror — it never reaches the network.
+   */
+  protected async fetchInternalFile(
+    src: string,
+  ): Promise<{ buffer: Buffer; mime: string }> {
+    const HARD_CEILING = 64 * 1024 * 1024; // 64 MiB memory guard
+    const relPath = resolveInternalFilePath(src);
+    const response = await this.client.get(relPath, {
+      responseType: "arraybuffer",
+      timeout: 30000,
+      maxContentLength: HARD_CEILING,
+      maxBodyLength: HARD_CEILING,
+    });
+    const buffer = Buffer.from(response.data);
+    if (buffer.length === 0) {
+      throw new Error(`Empty file response from "${src}"`);
+    }
+    const rawCt = response.headers?.["content-type"];
+    const mime =
+      typeof rawCt === "string" && rawCt.length > 0
+        ? rawCt.split(";")[0].trim().toLowerCase()
+        : "application/octet-stream";
+    return { buffer, mime };
+  }
+
+  /**
+   * Stash a page's full content into the in-RAM blob sandbox and return ONLY a
+   * short anonymous URL — the body never enters the model context (this is the
+   * whole point: ~30KB+ ProseMirror docs blow the model context if passed as a
+   * tool argument). Every INTERNAL file/image src (the type-agnostic criterion,
+   * so drawio/excalidraw/video/file nodes are covered too) is mirrored into the
+   * sandbox and its `src` rewritten to the sandbox URL, so an external consumer
+   * can fetch the images anonymously. External http(s) srcs are left untouched.
+   *
+   * Blobs live in RAM with a short TTL and are cleared on restart — consume the
+   * URLs within the TTL and one uptime. A failed image fetch never aborts the
+   * doc: the original src is kept and the failure counted.
+   *
+   * Returns { uri, sha256, size, images:{mirrored, failed} }. `uri` and `sha256`
+   * are for the document blob; `sha256` is also the blob's ETag (integrity).
+   */
+  async stashPage(pageId: string): Promise<{
+    uri: string;
+    sha256: string;
+    size: number;
+    images: { mirrored: number; failed: number };
+  }> {
+    if (!this.sandboxPut) {
+      throw new Error(
+        "stashPage is unavailable: the blob sandbox is not configured on this server",
+      );
+    }
+    await this.ensureAuthenticated();
+
+    // Stash the SAME shape getPageJson returns (id/title/.../content), with a
+    // deep clone so the rewrite never mutates anything shared.
+    const pageJson = await this.getPageJson(pageId);
+    const cloned: any = structuredClone(pageJson);
+
+    // Group internal-file nodes by normalized src so each unique resource is
+    // fetched + stored ONCE (dedup), and every node sharing that src points at
+    // the one sandbox blob. Capture each node's ORIGINAL raw src per-node:
+    // dedup groups nodes whose normalized src is equal even when their raw srcs
+    // differ (e.g. `/api/files/...` vs the bare `/files/...`), so on a revert we
+    // must restore each node's own original value, not the group key.
+    const bySrc = new Map<string, Array<{ node: any; origSrc: string }>>();
+    for (const node of collectInternalFileNodes(cloned.content)) {
+      const origSrc = String(node.attrs.src);
+      const src = normalizeFileUrl(origSrc);
+      const entry = { node, origSrc };
+      const group = bySrc.get(src);
+      if (group) group.push(entry);
+      else bySrc.set(src, [entry]);
+    }
+
+    let mirrored = 0;
+    let failed = 0;
+    // Record every successful mirror so it can be (a) reverted if its blob gets
+    // FIFO-evicted by a LATER put in this same stash, and (b) freed if the final
+    // doc put throws.
+    const mirrors: Array<{
+      uri: string;
+      entries: Array<{ node: any; origSrc: string }>;
+    }> = [];
+    const MAX_CONCURRENCY = 5;
+    const groups = [...bySrc.entries()];
+    for (let i = 0; i < groups.length; i += MAX_CONCURRENCY) {
+      const batch = groups.slice(i, i + MAX_CONCURRENCY);
+      await Promise.all(
+        batch.map(async ([src, entries]) => {
+          try {
+            const { buffer, mime } = await this.fetchInternalFile(src);
+            // put may throw if the blob exceeds the per-blob/total caps.
+            const stored = this.sandboxPut!(buffer, mime);
+            for (const entry of entries) entry.node.attrs.src = stored.uri;
+            mirrors.push({ uri: stored.uri, entries });
+            mirrored++;
+          } catch (err) {
+            // One bad/oversized image (or a rejected traversal src) must not
+            // abort the document. Logged unconditionally (never the blob body),
+            // matching the package's ungated console.warn convention.
+            failed++;
+            console.warn(
+              `stashPage: failed to mirror "${src}": ${
+                err instanceof Error ? err.message : String(err)
+              }`,
+            );
+          }
+        }),
+      );
+    }
+
+    // Revert one mirror's nodes to their original internal srcs and re-count it
+    // as failed (its blob was FIFO-evicted before the doc could reference it
+    // safely).
+    const revertMirror = (mirror: {
+      uri: string;
+      entries: Array<{ node: any; origSrc: string }>;
+    }) => {
+      for (const entry of mirror.entries) entry.node.attrs.src = entry.origSrc;
+      mirrored--;
+      failed++;
+      console.warn(
+        `stashPage: mirrored blob ${mirror.uri} was evicted before the doc ` +
+          `could safely reference it; reverted its src and counted it as failed`,
+      );
+    };
+
+    // Pre-put reconciliation: an image put earlier in THIS stash can FIFO-evict
+    // an even-earlier image of the same stash. Drop those from the live set
+    // first so the first serialized doc is already mostly correct.
+    let liveMirrors = mirrors;
+    if (this.sandboxHas) {
+      liveMirrors = [];
+      for (const mirror of mirrors) {
+        if (this.sandboxHas(mirror.uri)) liveMirrors.push(mirror);
+        else revertMirror(mirror);
+      }
+    }
+
+    // Put the document, then reconcile against eviction caused by the doc put
+    // ITSELF (the doc is newest, FIFO drops oldest = this stash's images). Each
+    // iteration reverts >=1 mirror, so the loop terminates (worst case: all
+    // images reverted and the doc references no sandbox image URLs).
+    let stored: { uri: string; sha256: string; size: number };
+    for (;;) {
+      const docBuf = Buffer.from(JSON.stringify(cloned), "utf8");
+      let docStored: { uri: string; sha256: string; size: number };
+      try {
+        docStored = this.sandboxPut(docBuf, "application/json");
+      } catch (err) {
+        // The doc put failed (e.g. doc exceeds the cap). Free this op's image
+        // blobs instead of leaking them in RAM for the whole TTL, then
+        // re-throw.
+        if (this.sandboxEvict) {
+          for (const mirror of liveMirrors) this.sandboxEvict(mirror.uri);
+        }
+        throw err;
+      }
+
+      if (!this.sandboxHas) {
+        stored = docStored;
+        break;
+      }
+      const evictedNow = liveMirrors.filter((m) => !this.sandboxHas!(m.uri));
+      if (evictedNow.length === 0) {
+        stored = docStored;
+        break;
+      }
+      // The doc we just stored references now-dead blobs. Revert those nodes,
+      // drop the stale doc blob, and loop to re-serialize + re-put the
+      // corrected doc.
+      for (const mirror of evictedNow) revertMirror(mirror);
+      liveMirrors = liveMirrors.filter((m) => this.sandboxHas!(m.uri));
+      if (this.sandboxEvict) this.sandboxEvict(docStored.uri);
+    }
+    return {
+      uri: stored.uri,
+      sha256: stored.sha256,
+      size: stored.size,
+      images: { mirrored, failed },
+    };
+  }
+
+  /**
+   * Compact outline of a page's top-level blocks (no full document body).
+   * Cheap way to locate sections/tables and grab block ids before drilling in
+   * with getNode / patchNode / insertNode.
+   */
+  }
+  return StashMixin;
+}

packages/mcp/src/client/tables.ts

@@ -0,0 +1,291 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import {
+  updatePageContentRealtime,
+  replacePageContent,
+  markdownToProseMirror,
+  markdownToProseMirrorCanonical,
+  mutatePageContent,
+  assertYjsEncodable,
+  MutationResult,
+} from "../lib/collaboration.js";
+import {
+  replaceNodeById,
+  replaceNodeByIdWithMany,
+  reassignCollidingBlockIds,
+  deleteNodeById,
+  assertUnambiguousMatch,
+  insertNodeRelative,
+  insertNodesRelative,
+  blockPlainText,
+  buildOutline,
+  getNodeByRef,
+  readTable,
+  insertTableRow,
+  deleteTableRow,
+  updateTableCell,
+  findInvalidNode,
+} from "@docmost/prosemirror-markdown";
+import { withPageLock, isUuid } from "../lib/page-lock.js";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+
+// Public method surface of TablesMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements ITablesMixin` fails to compile on drift.
+export interface ITablesMixin {
+  insertFootnote(pageId: string, anchorText: string, text: string): any;
+  tableInsertRow(pageId: string, tableRef: string, cells: string[], index?: number): any;
+  tableDeleteRow(pageId: string, tableRef: string, index: number): any;
+  tableUpdateCell(pageId: string, tableRef: string, row: number, col: number, text: string): any;
+}
+
+export function TablesMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & ITablesMixin> & TBase {
+  abstract class TablesMixin extends Base implements ITablesMixin {
+  /**
+   * AUTHOR-INLINE footnote insertion. The agent supplies only WHERE
+   * (`anchorText`, a snippet of body text to attach the marker after) and WHAT
+   * (`text`, the footnote content as markdown). Numbering and the bottom
+   * `footnotesList` are derived deterministically server-side
+   * (`insertInlineFootnote` -> `canonicalizeFootnotes`): the agent never sees,
+   * assigns, or edits a footnote number or the list, so it CANNOT desync.
+   *
+   * Content DEDUP: when an existing definition has the same content, its id is
+   * reused (one number, one definition, several references). The write is atomic
+   * via `mutatePageContent` (single-writer, page-locked); if the anchor text is
+   * not found the transform aborts with a clear error and no write happens.
+   */
+  async insertFootnote(pageId: string, anchorText: string, text: string) {
+    await this.ensureAuthenticated();
+    if (!anchorText || !anchorText.trim()) {
+      throw new Error("insertFootnote: anchorText is required");
+    }
+    if (text == null || `${text}`.trim() === "") {
+      throw new Error("insertFootnote: text is required");
+    }
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+    let result: { footnoteId: string; reused: boolean } | null = null;
+    const mutation = await this.mutatePage(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc: any) => {
+        const r = insertInlineFootnote(liveDoc, { anchorText, text });
+        if (!r.inserted) {
+          // Abort the page-locked write by throwing: mutatePageContent does not
+          // persist when the transform throws, so a missing anchor leaves the
+          // page untouched (no partial write).
+          throw new Error(
+            `insertFootnote: anchor text not found: ${JSON.stringify(
+              anchorText.slice(0, 80),
+            )}`,
+          );
+        }
+        result = { footnoteId: r.footnoteId, reused: r.reused };
+        return r.doc;
+      },
+    );
+    // The not-found path throws inside the transform (aborting mutatePage), so by
+    // here `result` is always set.
+    const r = result!;
+    return {
+      success: true,
+      modified: true,
+      pageId,
+      footnoteId: r.footnoteId,
+      reused: r.reused,
+      message: r.reused
+        ? "Footnote inserted (reused an existing same-content definition)."
+        : "Footnote inserted.",
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Page-locked write seam over collaboration.mutatePageContent. Production just
+   * delegates; it exists as an overridable method so the insertFootnote wrapper
+   * (transform abort-on-not-found + response shaping) can be unit-tested without
+   * standing up a live Hocuspocus collab socket.
+   *
+   * SELF-RESOLVES the pageId to the canonical UUID (issue #449, "resolve-then-
+   * lock"): every write must lock and key its CollabSession by the UUID, never a
+   * raw slugId (#260). resolvePageId is cached/idempotent, so a caller that
+   * already resolved pays no extra round-trip; centralizing it here means a
+   * caller that reaches this seam with a raw slugId still locks correctly instead
+   * of silently splitting the mutex key. withPageLock also asserts the key is a
+   * UUID as a hard backstop.
+   */
+
+  /**
+   * Insert a row of plain-text cells into a table on the LIVE collab document.
+   * `tableRef` is `#<index>` or a block id inside the target table. `cells` is
+   * padded to the table's column count (more cells than columns throws); `index`
+   * is a 0-based insert position (omit/out-of-range to append). Throws when no
+   * table resolves for the reference.
+   */
+  async tableInsertRow(
+    pageId: string,
+    tableRef: string,
+    cells: string[],
+    index?: number,
+  ) {
+    await this.ensureAuthenticated();
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    // Track insertion in an outer var, reset per-transform, so a collab retry
+    // recomputes it cleanly (mirrors insertNode's pattern).
+    let inserted = false;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        inserted = false;
+        const { doc: nd, inserted: ins } = insertTableRow(
+          liveDoc,
+          tableRef,
+          cells,
+          index,
+        );
+        inserted = ins;
+        if (!inserted) return null; // table not found -> skip the write entirely
+        return nd;
+      },
+    );
+
+    if (!inserted) {
+      throw new Error(
+        `tableInsertRow: no table found for "${tableRef}" on page ${pageId} (use "#<index>" from getOutline, or a block id inside the table)`,
+      );
+    }
+    return {
+      success: true,
+      table: tableRef,
+      inserted: true,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Delete the row at 0-based `index` from a table on the LIVE collab document.
+   * `tableRef` is `#<index>` or a block id inside the target table. The helper's
+   * out-of-range and last-row errors propagate; a missing table throws here.
+   */
+  async tableDeleteRow(pageId: string, tableRef: string, index: number) {
+    await this.ensureAuthenticated();
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    let deleted = false;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        deleted = false;
+        const { doc: nd, deleted: del } = deleteTableRow(
+          liveDoc,
+          tableRef,
+          index,
+        );
+        deleted = del;
+        if (!deleted) return null; // table not found -> skip the write entirely
+        return nd;
+      },
+    );
+
+    if (!deleted) {
+      throw new Error(
+        `tableDeleteRow: no table found for "${tableRef}" on page ${pageId} (use "#<index>" from getOutline, or a block id inside the table)`,
+      );
+    }
+    return {
+      success: true,
+      table: tableRef,
+      deleted: true,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Set the plain-text content of cell `[row, col]` (0-based) in a table on the
+   * LIVE collab document, replacing the cell's content with a single text
+   * paragraph (the cell's first-paragraph id is preserved). `tableRef` is
+   * `#<index>` or a block id inside the target table. The helper's out-of-range
+   * error propagates; a missing table throws here.
+   */
+  async tableUpdateCell(
+    pageId: string,
+    tableRef: string,
+    row: number,
+    col: number,
+    text: string,
+  ) {
+    await this.ensureAuthenticated();
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+
+    let updated = false;
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      (liveDoc) => {
+        updated = false;
+        const { doc: nd, updated: upd } = updateTableCell(
+          liveDoc,
+          tableRef,
+          row,
+          col,
+          text,
+        );
+        updated = upd;
+        if (!updated) return null; // table not found -> skip the write entirely
+        return nd;
+      },
+    );
+
+    if (!updated) {
+      throw new Error(
+        `tableUpdateCell: no table found for "${tableRef}" on page ${pageId} (use "#<index>" from getOutline, or a block id inside the table)`,
+      );
+    }
+    return {
+      success: true,
+      table: tableRef,
+      row,
+      col,
+      verify: mutation.verify,
+    };
+  }
+
+  /**
+   * Create a new page with title and content.
+   * Uses the /pages/import workaround (the only endpoint accepting content),
+   * then moves the page and restores the exact title: the import endpoint
+   * derives the title from the FILENAME and replaces spaces with
+   * underscores, so we explicitly re-set it via /pages/update afterwards.
+   */
+  }
+  return TablesMixin;
+}

packages/mcp/src/client/transforms.ts

@@ -0,0 +1,232 @@
+// Auto-split from client.ts (issue #450). Mixin over the shared client context.
+// Bodies are VERBATIM from the original DocmostClient; only the enclosing class
+// changed to a mixin factory. See client/context.ts for the shared base.
+import type { GConstructor, DocmostClientContext } from "./context.js";
+import {
+  updatePageContentRealtime,
+  replacePageContent,
+  markdownToProseMirror,
+  markdownToProseMirrorCanonical,
+  mutatePageContent,
+  assertYjsEncodable,
+  MutationResult,
+} from "../lib/collaboration.js";
+import { diffDocs, summarizeChange } from "../lib/diff.js";
+import {
+  blockText,
+  walk,
+  getList,
+  insertMarkerAfter,
+  setCalloutRange,
+  noteItem,
+  mdToInlineNodes,
+  commentsToFootnotes,
+  canonicalizeFootnotes,
+  insertInlineFootnote,
+  mergeFootnoteDefinitions,
+} from "../lib/transforms.js";
+import { normalizeAndMergeFootnotes } from "../lib/footnote-normalize-merge.js";
+import vm from "node:vm";
+
+// Public method surface of TransformsMixin (issue #450) — a NAMED type so the factory
+// return type is expressible in the emitted .d.ts (the anonymous mixin class
+// carries the base's protected shared state, which would otherwise trip TS4094).
+// Derived from the class below; `implements ITransformsMixin` fails to compile on drift.
+export interface ITransformsMixin {
+  transformPage(pageId: string, transformJs: string, opts?: { dryRun?: boolean; deleteComments?: boolean }): any;
+}
+
+export function TransformsMixin<TBase extends GConstructor<DocmostClientContext>>(Base: TBase): GConstructor<DocmostClientContext & ITransformsMixin> & TBase {
+  abstract class TransformsMixin extends Base implements ITransformsMixin {
+  /**
+   * Edit a page by running an arbitrary user-supplied JS transform against the
+   * live document, with a diff preview + page-history safety net.
+   *
+   * The transform string is evaluated as `(doc, ctx) => doc` inside a node:vm
+   * sandbox: it gets ONLY `{ doc, ctx, structuredClone, console }` as globals,
+   * a 5s timeout, and NO access to require/process/fs/network. It must return a
+   * `{ type: "doc" }` node, which is validated structurally before any write.
+   *
+   * `ctx` exposes:
+   *   - comments: the page's comments (fetched before the live read);
+   *   - log: an array the transform can push diagnostics to (via console.log);
+   *   - consume(id): mark a comment id as consumed (for deleteComments);
+   *   - helpers: the transforms.ts primitives + commentsToFootnotes.
+   *
+   * Footnote convention used by the helpers: footnote markers are plain "[N]"
+   * text in the body, and the notes are an orderedList under a heading whose
+   * text is "Примечания переводчика".
+   *
+   * dryRun (default true): read the page's current content, run the transform,
+   * and return `{ pushed:false, diff, log }` WITHOUT opening the collab socket.
+   * Otherwise the transform runs atomically inside mutatePageContent, optionally
+   * deletes consumed comments, and returns the new historyId + diff + log.
+   */
+  async transformPage(
+    pageId: string,
+    transformJs: string,
+    opts: { dryRun?: boolean; deleteComments?: boolean } = {},
+  ) {
+    const dryRun = opts.dryRun ?? true;
+    const deleteComments = opts.deleteComments ?? false;
+
+    await this.ensureAuthenticated();
+    // Full feed (incl. resolved): a page transform (e.g. comments -> footnotes)
+    // must operate on every comment, so it opts into the unfiltered feed.
+    const comments = (await this.listComments(pageId, true)).items;
+
+    // ctx handed to the sandbox. consume() records ids; helpers are the pure
+    // transform primitives. log is captured from console.log inside the sandbox.
+    const ctx = {
+      comments,
+      log: [] as string[],
+      consumed: new Set<string>(),
+      consume(id: string) {
+        this.consumed.add(id);
+      },
+      helpers: {
+        blockText,
+        walk,
+        getList,
+        insertMarkerAfter,
+        setCalloutRange,
+        noteItem,
+        mdToInlineNodes,
+        commentsToFootnotes,
+        canonicalizeFootnotes,
+        insertInlineFootnote,
+      },
+    };
+
+    // Captured oldDoc / newDoc for the diff (set inside runTransform).
+    let oldDoc: any;
+    let newDoc: any;
+
+    // SYNCHRONOUS transform runner — safe to call inside mutatePageContent's
+    // onSynced (no await between the live read and the write).
+    const runTransform = (liveDoc: any): any => {
+      oldDoc = structuredClone(liveDoc);
+      const sandbox: Record<string, any> = {
+        doc: structuredClone(liveDoc),
+        ctx,
+        structuredClone,
+        console: {
+          log: (...a: any[]) => ctx.log.push(a.map((x) => String(x)).join(" ")),
+        },
+      };
+      // Wrap the provided string in parentheses so both an expression-arrow
+      // (`(doc, ctx) => {...}`) and a parenthesized function work. Run it in a
+      // fresh context with no require/process/module so the transform cannot
+      // touch fs/network/process. 5s wall-clock timeout.
+      let fn: any;
+      try {
+        fn = vm.runInNewContext("(" + transformJs + ")", sandbox, {
+          timeout: 5000,
+        });
+      } catch (e: any) {
+        throw new Error(`transform did not compile: ${e?.message ?? e}`);
+      }
+      if (typeof fn !== "function") {
+        throw new Error(
+          "transform must evaluate to a function (doc, ctx) => doc",
+        );
+      }
+      const raw = vm.runInNewContext(
+        "f(d, c)",
+        { f: fn, d: sandbox.doc, c: ctx },
+        { timeout: 5000 },
+      );
+      if (
+        !raw ||
+        typeof raw !== "object" ||
+        raw.type !== "doc" ||
+        !Array.isArray(raw.content)
+      ) {
+        throw new Error(
+          'transform must return a ProseMirror doc node ({ type:"doc", content:[...] })',
+        );
+      }
+      // Validate the RAW transform output FIRST (structure — including the
+      // MAX_DEPTH guard — and URLs), mirroring updatePageJson. The canonicalizer
+      // recurses without a depth limiter, so validating after it would turn a
+      // too-deep doc into an opaque "Maximum call stack size exceeded" instead of
+      // the intended "nesting exceeds the maximum depth" error.
+      this.validateDocStructure(raw);
+      this.validateDocUrls(raw);
+      // Auto-canonicalize footnotes after the transform (idempotent): no write
+      // path can leave footnotes out of order / orphaned / in a raw `[^id]`
+      // block. In a dryRun preview this may surface footnote edits the script
+      // author did not write (the canonicalizer tidied them) — that is expected.
+      // #419: normalize + merge glyph-forked definitions before canonicalizing.
+      const result = canonicalizeFootnotes(normalizeAndMergeFootnotes(raw));
+      newDoc = result;
+      return result;
+    };
+
+    if (dryRun) {
+      // Preview only: run against the current REST snapshot, never open the
+      // socket. oldDoc/newDoc are captured by runTransform.
+      const raw = await this.getPageRaw(pageId);
+      const current = raw.content || { type: "doc", content: [] };
+      runTransform(current);
+      // Run an independent Yjs-encodability check (same sanitize + schema as the
+      // apply path), so the preview fails with the same descriptive error when
+      // the doc is not encodable instead of returning a misleadingly-green diff.
+      assertYjsEncodable(newDoc);
+      return {
+        pushed: false,
+        diff: diffDocs(oldDoc, newDoc),
+        log: ctx.log,
+      };
+    }
+
+    // Apply atomically against the live doc.
+    const collabToken = await this.getCollabTokenWithReauth();
+    // Open the collab doc by the canonical UUID, never the slugId (#260).
+    const pageUuid = await this.resolvePageId(pageId);
+    const mutation = await mutatePageContent(
+      pageUuid,
+      collabToken,
+      this.apiUrl,
+      runTransform,
+    );
+
+    // Optionally delete consumed comments (best-effort; a delete failure must
+    // not undo the successful write).
+    const deletedComments: string[] = [];
+    if (deleteComments) {
+      for (const id of ctx.consumed) {
+        try {
+          await this.deleteComment(id);
+          deletedComments.push(id);
+        } catch (e) {
+          if (process.env.DEBUG) {
+            console.error(`transform: failed to delete comment ${id}:`, e);
+          }
+        }
+      }
+    }
+
+    // Fetch the newest historyId (Docmost snapshots on the write above).
+    let historyId: string | null = null;
+    try {
+      const hist = await this.listPageHistory(pageId);
+      historyId = hist.items?.[0]?.id ?? null;
+    } catch (e) {
+      if (process.env.DEBUG) {
+        console.error("transform: failed to fetch history id:", e);
+      }
+    }
+
+    return {
+      pushed: true,
+      historyId,
+      diff: diffDocs(oldDoc, newDoc),
+      deletedComments,
+      log: ctx.log,
+      verify: mutation.verify,
+    };
+  }
+  }
+  return TransformsMixin;
+}

packages/mcp/src/index.ts

@@ -435,30 +435,63 @@ server.registerTool(
 // Tool: search
 // INTENTIONAL per-transport divergence (not shared): the in-app `searchPages`
 // runs a semantic + keyword hybrid (RRF) with in-process access control and a
-// different schema (limit 1-20); this transport is a plain REST full-text search
+// different schema; this transport is the #443 agent-lookup search — a hybrid
-// (limit up to 100). Different behaviour AND schema, so kept per-layer.
+// substring + full-text search that also returns each hit's location (`path`)
+// and a windowed `snippet`, so one call answers "where is it and what's in it".
+// The in-app hybrid-RRF search is deliberately NOT touched. Different behaviour
+// AND schema, so kept per-layer.
+//
+// STANDALONE-vs-STOCK-UPSTREAM: the client sends the opt-in `substring`/
+// `parentPageId`/`titleOnly` DTO fields. A stock upstream server validates the
+// DTO with `whitelist: true` and silently strips these unknown fields, so the
+// request degrades gracefully to plain FTS (no path/snippet, current shape).
+//
+// EE/TYPESENSE DEGRADATION (#443): on an instance whose SEARCH_DRIVER is
+// `typesense`, the server routes this request to the Typesense backend, which
+// does NOT implement agent-lookup — the substring/path/snippet/tiering is
+// ignored and the response degrades to plain Typesense FTS. The rich lookup
+// shape is only produced by the native Postgres search driver.
 server.registerTool(
   "search",
   {
     description:
-      "Full-text search for pages and content across the whole workspace. " +
+      "Find pages by a fragment of a technical string (hostnames, IPs, IDs " +
-      "Results are bounded by `limit` (1-100; when omitted the server applies " +
+      "like `srv.local`, `10.0.12`, `WB-MGE-30D86B`) — one call returns each " +
-      "its own default).",
+      "hit's location (`path`: ancestor titles root→parent) and a `snippet` " +
+      "around the first match, so you rarely need a follow-up get_page. " +
+      "Matches substrings literally (dots/dashes/digits are not tokenized) as " +
+      "well as full-text. Returns `{ pageId, title, path, snippet, score }` " +
+      "sorted by `score` (a per-response relevance float).",
     inputSchema: {
       query: z.string().min(1).describe("Search query"),
+      spaceId: z
+        .string()
+        .optional()
+        .describe("Restrict the search to a single space"),
+      parentPageId: z
+        .string()
+        .optional()
+        .describe(
+          "Restrict to a page and all its descendants (the page itself included)",
+        ),
+      titleOnly: z
+        .boolean()
+        .optional()
+        .describe("Match page titles only; skip page text"),
       limit: z
         .number()
         .int()
         .min(1)
-        .max(100)
+        .max(50)
         .optional()
-        .describe("Max results to return (max 100)"),
+        .describe("Max results to return (1-50, default 10)"),
     },
   },
-  async ({ query, limit }) => {
+  async ({ query, spaceId, parentPageId, titleOnly, limit }) => {
-    // The tool exposes no spaceId filter, so pass undefined for the client's
+    const result = await docmostClient.search(query, spaceId, limit, {
-    // optional spaceId parameter and forward limit into its correct slot.
+      parentPageId,
-    const result = await docmostClient.search(query, undefined, limit);
+      titleOnly,
+    });
     return jsonContent(result);
   },
 );

packages/mcp/src/lib/collab-session.ts

@@ -440,9 +440,16 @@ export class CollabSession {
       // must stay synchronous (no await). While the JS event loop is not
       // yielded, no incoming remote update can interleave, so any already-synced
       // concurrent edits are preserved in liveDoc.
+      //
+      // INVARIANT 1 is machine-checked: the BEGIN/END markers below delimit the
+      // no-await window, and test/unit/no-await-critical-window.test.mjs scans
+      // this source and FAILS if any `await` (or `for await`/`yield`) appears
+      // between them. Do NOT add an await inside this block — an accidental
+      // async boundary here silently reopens the clobber-live-edits race (#152).
       let newDoc: any;
       let beforeDoc: any;
       try {
+        // === MUTATE-CRITICAL-WINDOW: BEGIN (no await between here and END #449) ===
         let liveDoc = TiptapTransformer.fromYdoc(this.ydoc, "default");
         if (
           !liveDoc ||
@@ -480,6 +487,7 @@ export class CollabSession {
         // ids of unchanged nodes, so an open editor's cursor is not yanked to the
         // end of the document on every agent write.
         applyDocToFragment(this.ydoc, newDoc);
+        // === MUTATE-CRITICAL-WINDOW: END (#449) ===
       } catch (e) {
         // Includes errors thrown by transform (e.g. "afterText not found",
         // "text not found"): propagate them verbatim to the caller.

packages/mcp/src/lib/diff.ts

@@ -14,7 +14,19 @@
  * signature.
  *
  * If recreateTransform / the changeset throws on a pathological document pair,
- * we fall back to a coarse block-level text diff so the tool never hard-fails.
+ * OR the pair is too large to diff cheaply (see the size guard below), we fall
+ * back to a coarse block-level text diff so the tool never hard-fails and never
+ * pins the event loop.
+ *
+ * SIZE GUARD (issue #464 — prod CPU-DoS). recreateTransform computes its diff via
+ * rfc6902.createPatch, whose array diff is O(n·m) Levenshtein per array pair and
+ * whose per-run word diff is O(w²); on a large/heavily-changed doc this runs for
+ * seconds-to-hours and starves the whole process (BullMQ, Redis lock renewals,
+ * embeddings). It never THROWS — it just never finishes — so the try/catch below
+ * cannot save us. Because diffDocs runs on EVERY in-app/MCP content edit's verify
+ * report, we PRE-FLIGHT the doc size and route anything above a cheap cap straight
+ * to the coarse fallback (the same shape the catch produces). Same cap+fallback
+ * pattern as the ELK-layout DoS fix (#440 / c917dcc3).
  */
 
 import { Node } from "@tiptap/pm/model";
@@ -72,6 +84,56 @@ function countNodes(doc: any, pred: (node: any) => boolean): number {
   return n;
 }
 
+// --- Issue #464: pre-flight size guard for the precise diff ------------------
+// Defaults are BENCHMARK-derived on the recreateTransform(complexSteps:false,
+// wordDiffs:true, simplifyDiff:true) pipeline, chosen so the WORST case (a fully
+// re-written doc — the adversarial shape that drove the incident) keeps the
+// synchronous block under ~200ms REGARDLESS of input:
+//   - 150 total nodes: worst-case pair ~176ms; the O(node²) array diff crosses
+//     200ms at ~170 nodes and then explodes super-linearly (400 nodes ~1.3s,
+//     800 ~5.5s), so cap just below the crossover.
+//   - 12 KiB serialized JSON: an independent axis, because the per-run word diff
+//     is O(words²) — a FEW nodes with very long text runs is dangerous even at a
+//     low node count (17 nodes / ~11 KiB ~176ms, / ~14 KiB ~290ms). A node-light
+//     but byte-heavy doc is still refused.
+// Either metric over its cap routes to the coarse fallback. Both are env-tunable
+// for operators who accept more CPU in exchange for exact diffs on larger docs.
+const DEFAULT_MAX_NODES = 150;
+const DEFAULT_MAX_BYTES = 12 * 1024;
+
+/**
+ * Read a positive-integer env override, falling back to `dflt`. Garbage / unset /
+ * non-finite / non-positive all fall back (so the guard can never be accidentally
+ * disabled by a malformed value). Read fresh on every call so a test / operator
+ * can flip the knob without a restart.
+ */
+function readPositiveIntEnv(name: string, dflt: number): number {
+  const raw = parseInt(process.env[name] ?? "", 10);
+  return Number.isFinite(raw) && raw > 0 ? raw : dflt;
+}
+
+/**
+ * True when the pair is too large for the precise (recreateTransform) diff and
+ * must degrade to the coarse fallback. Takes the MAX of the two docs on each
+ * metric so an ASYMMETRIC pair (a small new doc vs a huge old doc, or vice
+ * versa) — which still explodes rfc6902 — is caught. Cheap: one node walk +
+ * one JSON.stringify per doc, both O(size).
+ */
+function exceedsDiffSizeGuard(oldDoc: any, newDoc: any): boolean {
+  const maxNodes = readPositiveIntEnv("MCP_DIFF_MAX_NODES", DEFAULT_MAX_NODES);
+  const maxBytes = readPositiveIntEnv("MCP_DIFF_MAX_BYTES", DEFAULT_MAX_BYTES);
+  const nodes = Math.max(
+    countNodes(oldDoc, () => true),
+    countNodes(newDoc, () => true),
+  );
+  if (nodes > maxNodes) return true;
+  const bytes = Math.max(
+    JSON.stringify(oldDoc)?.length ?? 0,
+    JSON.stringify(newDoc)?.length ?? 0,
+  );
+  return bytes > maxBytes;
+}
+
 /**
  * Count UNIQUE links in a JSON doc by their `href`. A single link can be split
  * across several adjacent text runs (e.g. a "link+bold" run followed by a "link"
@@ -226,6 +288,81 @@ function coarseDiff(oldDoc: any, newDoc: any): DiffChange[] {
   return changes;
 }
 
+/** Accumulated textual changes plus their derived char/block tallies. */
+interface DiffTally {
+  changes: DiffChange[];
+  inserted: number;
+  deleted: number;
+  changedBlocks: Set<string>;
+}
+
+/**
+ * Produce the coarse-fallback tally for a pair. This is the SINGLE source of the
+ * `fellBack:true` result shape, shared by BOTH degrade paths in diffDocs (the
+ * pre-flight size guard and the recreateTransform catch) so they behave and
+ * report identically.
+ */
+function coarseDiffTally(oldDoc: any, newDoc: any): DiffTally {
+  const changes = coarseDiff(oldDoc, newDoc);
+  let inserted = 0;
+  let deleted = 0;
+  const changedBlocks = new Set<string>();
+  for (const c of changes) {
+    if (c.op === "insert") inserted += c.text.length;
+    else deleted += c.text.length;
+    if (c.block) changedBlocks.add(c.op[0] + ":" + c.block);
+  }
+  return { changes, inserted, deleted, changedBlocks };
+}
+
+/**
+ * Compute the PRECISE tally via the recreateTransform pipeline. Callers MUST
+ * gate this behind the size guard (it can block the event loop for a large pair)
+ * and wrap it in try/catch (a pathological pair can throw); on either the guard
+ * or a throw, use `coarseDiffTally` instead. Kept as a sibling of
+ * `coarseDiffTally` so both produce the same `DiffTally` shape.
+ */
+function preciseDiffTally(oldDocJson: any, newDocJson: any): DiffTally {
+  const oldNode = Node.fromJSON(docmostSchema, oldDocJson);
+  const newNode = Node.fromJSON(docmostSchema, newDocJson);
+  const tr = recreateTransform(oldNode, newNode, {
+    complexSteps: false,
+    wordDiffs: true,
+    simplifyDiff: true,
+  });
+  const changeSet = ChangeSet.create(oldNode).addSteps(tr.doc, tr.mapping.maps, []);
+  const simplified = simplifyChanges(changeSet.changes, newNode);
+
+  const changes: DiffChange[] = [];
+  let inserted = 0;
+  let deleted = 0;
+  const changedBlocks = new Set<string>();
+
+  for (const change of simplified) {
+    // Deleted text lives in the OLD doc coordinate range [fromA, toA).
+    if (change.toA > change.fromA) {
+      const text = oldNode.textBetween(change.fromA, change.toA, "\n", " ");
+      if (text.length > 0) {
+        deleted += text.length;
+        const block = blockContextAt(oldNode, change.fromA);
+        changes.push({ op: "delete", block, text });
+        if (block) changedBlocks.add("d:" + block);
+      }
+    }
+    // Inserted text lives in the NEW doc coordinate range [fromB, toB).
+    if (change.toB > change.fromB) {
+      const text = newNode.textBetween(change.fromB, change.toB, "\n", " ");
+      if (text.length > 0) {
+        inserted += text.length;
+        const block = blockContextAt(newNode, change.fromB);
+        changes.push({ op: "insert", block, text });
+        if (block) changedBlocks.add("i:" + block);
+      }
+    }
+  }
+  return { changes, inserted, deleted, changedBlocks };
+}
+
 /** Build the human-readable unified-ish markdown summary. */
 function renderMarkdown(
   result: Omit<DiffResult, "markdown">,
@@ -276,66 +413,39 @@ export function diffDocs(
   newDocJson: any,
   notesHeading: string = "Примечания переводчика",
 ): DiffResult {
+  // computeIntegrity is cheap (linear node walks) and its counts are needed in
+  // BOTH the precise and coarse paths, so it always runs first.
   const integrity = computeIntegrity(oldDocJson, newDocJson, notesHeading);
 
-  let changes: DiffChange[] = [];
-  let inserted = 0;
-  let deleted = 0;
   let fellBack = false;
-  const changedBlocks = new Set<string>();
+  let tally: DiffTally;
 
-  try {
+  // Pre-flight size guard (#464): a too-large pair would make recreateTransform
-    const oldNode = Node.fromJSON(docmostSchema, oldDocJson);
+  // block the event loop for seconds-to-hours WITHOUT throwing, so route it to
-    const newNode = Node.fromJSON(docmostSchema, newDocJson);
+  // the coarse fallback BEFORE calling recreateTransform at all. Both this path
-    const tr = recreateTransform(oldNode, newNode, {
+  // and the catch below go through coarseDiffTally for an identical `fellBack`
-      complexSteps: false,
+  // result shape.
-      wordDiffs: true,
+  if (exceedsDiffSizeGuard(oldDocJson, newDocJson)) {
-      simplifyDiff: true,
-    });
-    const changeSet = ChangeSet.create(oldNode).addSteps(
-      tr.doc,
-      tr.mapping.maps,
-      [],
-    );
-    const simplified = simplifyChanges(changeSet.changes, newNode);
-
-    for (const change of simplified) {
-      // Deleted text lives in the OLD doc coordinate range [fromA, toA).
-      if (change.toA > change.fromA) {
-        const text = oldNode.textBetween(change.fromA, change.toA, "\n", " ");
-        if (text.length > 0) {
-          deleted += text.length;
-          const block = blockContextAt(oldNode, change.fromA);
-          changes.push({ op: "delete", block, text });
-          if (block) changedBlocks.add("d:" + block);
-        }
-      }
-      // Inserted text lives in the NEW doc coordinate range [fromB, toB).
-      if (change.toB > change.fromB) {
-        const text = newNode.textBetween(change.fromB, change.toB, "\n", " ");
-        if (text.length > 0) {
-          inserted += text.length;
-          const block = blockContextAt(newNode, change.fromB);
-          changes.push({ op: "insert", block, text });
-          if (block) changedBlocks.add("i:" + block);
-        }
-      }
-    }
-  } catch {
-    // Pathological pair: degrade to a coarse block-level diff so we never throw.
     fellBack = true;
-    changes = coarseDiff(oldDocJson, newDocJson);
+    tally = coarseDiffTally(oldDocJson, newDocJson);
-    for (const c of changes) {
+  } else {
-      if (c.op === "insert") inserted += c.text.length;
+    try {
-      else deleted += c.text.length;
+      tally = preciseDiffTally(oldDocJson, newDocJson);
-      if (c.block) changedBlocks.add(c.op[0] + ":" + c.block);
+    } catch {
+      // Pathological pair: degrade to a coarse block-level diff so we never throw.
+      fellBack = true;
+      tally = coarseDiffTally(oldDocJson, newDocJson);
     }
   }
 
   const partial: Omit<DiffResult, "markdown"> = {
-    summary: { inserted, deleted, blocksChanged: changedBlocks.size },
+    summary: {
+      inserted: tally.inserted,
+      deleted: tally.deleted,
+      blocksChanged: tally.changedBlocks.size,
+    },
     integrity,
-    changes,
+    changes: tally.changes,
   };
   return { ...partial, markdown: renderMarkdown(partial, fellBack) };
 }

packages/mcp/src/lib/drawio-cell-ops.ts

@@ -0,0 +1,168 @@
+// ID-based cell operations for `drawioEditCells` (issue #425, stage 3).
+//
+// Instead of resending the whole XML (whose diff is fragile — draw.io reorders
+// attributes, and a {search,replace} text match breaks on it), the model sends
+// targeted operations keyed by cell id:
+//
+//   { op: "add",    xml: "<mxCell .../>" }            // append a new cell
+//   { op: "update", cellId: "n3", xml: "<mxCell .../>" }  // replace that cell
+//   { op: "delete", cellId: "n5" }                    // + CASCADE
+//
+// `delete` CASCADES: it removes the cell, every descendant cell whose parent
+// chain leads to it (container children), AND every edge whose source or target
+// is any deleted cell. Ids are STABLE across edits so diffs stay meaningful.
+//
+// Operations apply to the parsed DOM of the current model; the caller re-lints
+// and rebuilds the .drawio.svg through the existing #423 pipeline afterwards.
+
+import { JSDOM } from "jsdom";
+
+let _window: any = null;
+function xmlWindow(): any {
+  if (!_window) _window = new JSDOM("").window;
+  return _window;
+}
+
+export type CellOp =
+  | { op: "add"; xml: string }
+  | { op: "update"; cellId: string; xml: string }
+  | { op: "delete"; cellId: string };
+
+export class CellOpsError extends Error {
+  constructor(message: string) {
+    super(`drawioEditCells: ${message}`);
+    this.name = "CellOpsError";
+  }
+}
+
+// The mxGraph root sentinels. id="0" is the graph root; id="1" is the default
+// layer that parents every real cell. A delete targeting either would cascade
+// through the whole diagram body (every cell chains up to "1"), so such an op is
+// rejected outright.
+const SENTINEL_IDS = new Set(["0", "1"]);
+
+/** Parse a single `<mxCell …>…</mxCell>` fragment into an element, or throw. */
+function parseCellFragment(xml: string): any {
+  const parser = new (xmlWindow().DOMParser)();
+  // Wrap so a self-closed or child-bearing single cell parses as one root.
+  const doc = parser.parseFromString(`<root>${xml}</root>`, "application/xml");
+  if (doc.getElementsByTagName("parsererror").length > 0) {
+    throw new CellOpsError(`operation xml is not well-formed: ${xml.slice(0, 120)}`);
+  }
+  const cells = doc.getElementsByTagName("mxCell");
+  if (cells.length !== 1) {
+    throw new CellOpsError(
+      `each add/update op must carry exactly one <mxCell> (got ${cells.length})`,
+    );
+  }
+  return cells[0];
+}
+
+/** All ids reachable as descendants of `rootId` via the parent relation. */
+function collectDescendants(
+  rootId: string,
+  parentOf: Map<string, string | undefined>,
+): Set<string> {
+  const doomed = new Set<string>([rootId]);
+  let grew = true;
+  while (grew) {
+    grew = false;
+    for (const [id, parent] of parentOf) {
+      if (!doomed.has(id) && parent != null && doomed.has(parent)) {
+        doomed.add(id);
+        grew = true;
+      }
+    }
+  }
+  return doomed;
+}
+
+/**
+ * Apply the operation list to a model XML string and return the new model XML.
+ * Uses the DOM so attribute order / formatting is preserved for untouched cells.
+ * Throws CellOpsError on an unknown target id or a malformed op fragment (so the
+ * model gets a precise error and nothing is half-applied).
+ */
+export function applyCellOps(modelXml: string, ops: CellOp[]): string {
+  if (!Array.isArray(ops) || ops.length === 0) {
+    throw new CellOpsError("operations must be a non-empty array");
+  }
+  const parser = new (xmlWindow().DOMParser)();
+  const doc = parser.parseFromString(modelXml, "application/xml");
+  if (doc.getElementsByTagName("parsererror").length > 0) {
+    throw new CellOpsError("the current diagram XML is not well-formed");
+  }
+  const root = doc.getElementsByTagName("root")[0];
+  if (!root) throw new CellOpsError("the current diagram has no <root> element");
+
+  const cellEls = () => Array.from(root.getElementsByTagName("mxCell")) as any[];
+  const byId = () => {
+    const m = new Map<string, any>();
+    for (const el of cellEls()) m.set(el.getAttribute("id") ?? "", el);
+    return m;
+  };
+
+  for (const op of ops) {
+    if (op.op === "add") {
+      const frag = parseCellFragment(op.xml);
+      const id = frag.getAttribute("id");
+      if (!id) throw new CellOpsError("an add op's <mxCell> is missing an id");
+      if (byId().has(id))
+        throw new CellOpsError(`add op id "${id}" already exists (use update)`);
+      root.appendChild(doc.importNode(frag, true));
+    } else if (op.op === "update") {
+      const map = byId();
+      const target = map.get(op.cellId);
+      if (!target)
+        throw new CellOpsError(`update target cell "${op.cellId}" does not exist`);
+      const frag = parseCellFragment(op.xml);
+      const newId = frag.getAttribute("id");
+      if (newId && newId !== op.cellId)
+        throw new CellOpsError(
+          `update op cellId "${op.cellId}" != the <mxCell> id "${newId}" (ids are stable)`,
+        );
+      // Replace the element in place so surrounding cells are untouched.
+      const imported = doc.importNode(frag, true);
+      target.parentNode.replaceChild(imported, target);
+    } else if (op.op === "delete") {
+      // Reject a sentinel-targeted delete BEFORE collecting descendants: "0"/"1"
+      // parent the entire diagram, so a cascade from either would wipe the whole
+      // model body (doomed.delete("0"/"1") only spared the sentinel itself, not
+      // its children).
+      if (SENTINEL_IDS.has(op.cellId))
+        throw new CellOpsError(
+          `cannot delete sentinel cell "${op.cellId}" (the graph root/default layer)`,
+        );
+      const map = byId();
+      if (!map.has(op.cellId))
+        throw new CellOpsError(`delete target cell "${op.cellId}" does not exist`);
+      // Build the parent relation over the CURRENT cells for the cascade.
+      const parentOf = new Map<string, string | undefined>();
+      for (const el of cellEls()) {
+        parentOf.set(el.getAttribute("id") ?? "", el.getAttribute("parent") ?? undefined);
+      }
+      const doomed = collectDescendants(op.cellId, parentOf);
+      // Cascade to edges whose source/target is any doomed cell.
+      for (const el of cellEls()) {
+        if (el.getAttribute("edge") !== "1") continue;
+        const src = el.getAttribute("source");
+        const tgt = el.getAttribute("target");
+        if ((src && doomed.has(src)) || (tgt && doomed.has(tgt))) {
+          doomed.add(el.getAttribute("id") ?? "");
+        }
+      }
+      // Never delete the sentinels even if referenced by a malformed op.
+      doomed.delete("0");
+      doomed.delete("1");
+      for (const el of cellEls()) {
+        const id = el.getAttribute("id") ?? "";
+        if (doomed.has(id)) el.parentNode.removeChild(el);
+      }
+    } else {
+      throw new CellOpsError(`unknown op "${(op as any).op}"`);
+    }
+  }
+
+  const ser = new (xmlWindow().XMLSerializer)();
+  return ser.serializeToString(doc.documentElement);
+}

packages/mcp/src/lib/drawio-graph.ts

@@ -0,0 +1,916 @@
+// Semantic graph -> draw.io pipeline for `drawioFromGraph` (issue #425, stage 3).
+//
+// The model describes a diagram SEMANTICALLY — nodes with a `kind` and an
+// optional `icon`, groups (containers), edges with a `kind` — and NEVER sees a
+// coordinate or a style string. This module owns the whole server-side pipeline:
+//
+//   1. validateGraph      — a hand-written validator (no zod dependency, so this
+//                           lib stays importable by client.ts without coupling to
+//                           a zod major) that rejects malformed graphs early.
+//   2. resolveNodeStyle   — `icon` -> exact style via the shape catalog (#424);
+//                           an UNKNOWN icon degrades to a generic shape by `kind`
+//                           WITH the label (never an empty square). `kind` -> the
+//                           preset palette slot.
+//   3. graphToElk         — graph -> ELK-JSON, honouring the layout hints
+//                           (`layer`/`sameLayerAs` -> layer constraints, `pinned`
+//                           -> a fixed node) and compound group nodes.
+//   4. assembleModel      — graph + ELK coordinates -> a full mxGraphModel XML
+//                           that satisfies the #423 linter BY CONSTRUCTION
+//                           (sentinels, transparent containers, relative child
+//                           coords, cross-container edges parent="1", >=150px
+//                           gaps from ELK spacing, escaped labels).
+//
+// The `layout` mode: "full" re-lays everything; "incremental" fixes existing
+// coordinates (ELK interactive mode) and places only new nodes; "none" keeps the
+// caller-provided/prior coordinates untouched.
+
+import ELK from "elkjs/lib/elk.bundled.js";
+import { JSDOM } from "jsdom";
+import {
+  searchShapes,
+  awsServiceStyle,
+  type ShapeResult,
+} from "./drawio-shapes.js";
+import {
+  getPreset,
+  genericNodeStyle,
+  iconNodeStyle,
+  edgeStyle,
+  groupStyle,
+  type PresetData,
+} from "./drawio-presets.js";
+import { MIN_SHAPE_GAP } from "./drawio-xml.js";
+
+// --- graph schema (plain TS + a hand validator) ----------------------------
+
+export interface GraphNode {
+  id: string;
+  label: string;
+  kind?: string;
+  /** Icon reference, e.g. "aws:lambda" | "azure:cosmos" | "lambda". */
+  icon?: string;
+  /** Group (container) id this node belongs to. */
+  group?: string;
+  /** Layer hint (ELK layerChoiceConstraint): 0-based column/row index. */
+  layer?: number;
+  /** Put this node in the same layer as another node id. */
+  sameLayerAs?: string;
+  /** Fix this node at exact coordinates (an ELK fixed node). */
+  pinned?: { x: number; y: number };
+}
+
+export interface GraphGroup {
+  id: string;
+  label: string;
+  kind?: string;
+  /** Parent group id — lets a group nest inside another group (e.g. subnet in VPC). */
+  group?: string;
+}
+
+export interface GraphEdge {
+  from: string;
+  to: string;
+  label?: string;
+  kind?: string;
+}
+
+export interface Graph {
+  nodes: GraphNode[];
+  groups?: GraphGroup[];
+  edges?: GraphEdge[];
+  direction?: "LR" | "RL" | "TB" | "BT";
+  preset?: string;
+}
+
+export type LayoutMode = "none" | "full" | "incremental";
+
+/** A structured validation error (mirrors the drawio linter's shape loosely). */
+export class GraphValidationError extends Error {
+  issues: string[];
+  constructor(issues: string[]) {
+    super(`drawioFromGraph: invalid graph — ${issues.join("; ")}`);
+    this.name = "GraphValidationError";
+    this.issues = issues;
+  }
+}
+
+const MAX_GRAPH_NODES = 500; // parity with drawio-layout's ELK_MAX_NODES.
+// Edge/group caps mirror drawio-layout's ELK_MAX_EDGES. Without an edge cap a
+// tiny node set with a huge edge list (e.g. 500 nodes / 200000 edges) passes
+// node validation, then graphToElk/runElk exhausts the heap SYNCHRONOUSLY inside
+// elk.bundled.js — before the 5s ELK timeout can fire and OUTSIDE it entirely
+// for the mapper/assembler — crashing the worker on LLM-authored input. Reject
+// the over-limit shape here, before any layout or assembly runs.
+export const MAX_GRAPH_EDGES = 1000; // parity with drawio-layout's ELK_MAX_EDGES.
+export const MAX_GRAPH_GROUPS = 500; // groups are compound ELK nodes; bound them too.
+
+/**
+ * Validate the graph structure BEFORE any layout/assembly so the model gets a
+ * precise, actionable error instead of a corrupt diagram. Throws
+ * GraphValidationError listing every problem.
+ */
+export function validateGraph(graph: Graph): void {
+  const issues: string[] = [];
+  if (!graph || typeof graph !== "object") {
+    throw new GraphValidationError(["graph must be an object"]);
+  }
+  if (!Array.isArray(graph.nodes) || graph.nodes.length === 0) {
+    throw new GraphValidationError(["graph.nodes must be a non-empty array"]);
+  }
+  // Size caps FIRST (fail fast, before touching per-element loops) so an
+  // over-limit graph can never reach the layout engine and OOM the worker.
+  if (graph.nodes.length > MAX_GRAPH_NODES) {
+    throw new GraphValidationError([
+      `graph has ${graph.nodes.length} nodes (max ${MAX_GRAPH_NODES})`,
+    ]);
+  }
+  if (Array.isArray(graph.edges) && graph.edges.length > MAX_GRAPH_EDGES) {
+    throw new GraphValidationError([
+      `graph has ${graph.edges.length} edges (max ${MAX_GRAPH_EDGES})`,
+    ]);
+  }
+  if (Array.isArray(graph.groups) && graph.groups.length > MAX_GRAPH_GROUPS) {
+    throw new GraphValidationError([
+      `graph has ${graph.groups.length} groups (max ${MAX_GRAPH_GROUPS})`,
+    ]);
+  }
+
+  const nodeIds = new Set<string>();
+  const groupIds = new Set<string>();
+  for (const g of graph.groups ?? []) {
+    if (!g.id) issues.push("a group is missing its id");
+    else if (groupIds.has(g.id)) issues.push(`duplicate group id "${g.id}"`);
+    groupIds.add(g.id);
+  }
+  for (const n of graph.nodes) {
+    if (!n.id) issues.push("a node is missing its id");
+    else if (nodeIds.has(n.id)) issues.push(`duplicate node id "${n.id}"`);
+    else if (groupIds.has(n.id))
+      issues.push(`node id "${n.id}" collides with a group id`);
+    nodeIds.add(n.id);
+    if (typeof n.label !== "string" || n.label === "")
+      issues.push(`node "${n.id}" is missing a label`);
+    if (n.group != null && !groupIds.has(n.group))
+      issues.push(`node "${n.id}" references unknown group "${n.group}"`);
+    if (n.pinned != null) {
+      if (
+        typeof n.pinned.x !== "number" ||
+        typeof n.pinned.y !== "number" ||
+        !Number.isFinite(n.pinned.x) ||
+        !Number.isFinite(n.pinned.y)
+      )
+        issues.push(`node "${n.id}" has an invalid pinned {x,y}`);
+    }
+    if (n.layer != null && (!Number.isInteger(n.layer) || n.layer < 0))
+      issues.push(`node "${n.id}" has an invalid layer (must be a >=0 integer)`);
+  }
+  // sameLayerAs must reference an existing node (checked after all ids known).
+  for (const n of graph.nodes) {
+    if (n.sameLayerAs != null && !nodeIds.has(n.sameLayerAs))
+      issues.push(
+        `node "${n.id}" sameLayerAs references unknown node "${n.sameLayerAs}"`,
+      );
+  }
+  for (const e of graph.edges ?? []) {
+    if (!e.from || !e.to) {
+      issues.push("an edge is missing from/to");
+      continue;
+    }
+    if (!nodeIds.has(e.from) && !groupIds.has(e.from))
+      issues.push(`edge from "${e.from}" resolves to no node/group`);
+    if (!nodeIds.has(e.to) && !groupIds.has(e.to))
+      issues.push(`edge to "${e.to}" resolves to no node/group`);
+  }
+  if (issues.length > 0) throw new GraphValidationError(issues);
+}
+
+// --- icon resolution -------------------------------------------------------
+
+/**
+ * Resolve a node's `icon` reference to a concrete style-string + size via the
+ * shape catalog. Accepts "aws:lambda", "azure:cosmos", or a bare "lambda". An
+ * AWS `resIcon` name is built directly (exact service-icon template). Anything
+ * else goes through searchShapes. Returns null when nothing resolves — the
+ * caller then falls back to a generic shape by kind (never an empty box).
+ */
+export function resolveIcon(icon: string): ShapeResult | null {
+  const raw = icon.trim();
+  if (raw === "") return null;
+  let provider = "";
+  let name = raw;
+  const colon = raw.indexOf(":");
+  if (colon !== -1) {
+    provider = raw.slice(0, colon).trim().toLowerCase();
+    name = raw.slice(colon + 1).trim();
+  }
+
+  if (provider === "aws") {
+    // Prefer an exact resIcon match from the catalog (carries the right size and
+    // any rebrand/blocklist note); if the underscore/space name doesn't hit,
+    // build the canonical service-icon style directly so it is never an empty box.
+    const results = searchShapes(name.replace(/_/g, " "), { limit: 5 });
+    const aws4 = results.find((r) => r.style.includes("mxgraph.aws4"));
+    if (aws4) return aws4;
+    return {
+      style: awsServiceStyle(name.replace(/\s+/g, "_")),
+      w: 78,
+      h: 78,
+      title: name,
+      type: "vertex",
+    };
+  }
+
+  // Non-AWS or bare name: fuzzy search the catalog. Take the top vertex hit,
+  // but REJECT a weak match (the fuzzy scorer can prefix-match an unrelated
+  // stencil, e.g. "not..." -> "Notebook"); require the hit's title to actually
+  // share a meaningful token with the query, otherwise degrade to generic-by-kind.
+  const q = provider ? `${provider} ${name}` : name;
+  const results = searchShapes(q, { limit: 8 });
+  const hit = results.find((r) => r.type !== "edge") ?? results[0];
+  if (!hit) return null;
+  if (!isRelevantMatch(name, hit.title)) return null;
+  return hit;
+}
+
+/**
+ * Whether a resolved stencil is a genuine match for the requested icon name (as
+ * opposed to a loose prefix hit on an unrelated shape). True if any 3+ char
+ * token of the query appears in the stencil title, or vice-versa.
+ */
+function isRelevantMatch(name: string, title: string): boolean {
+  const norm = (s: string) => s.toLowerCase().replace(/[^a-z0-9]+/g, " ").trim();
+  const qTokens = norm(name).split(/\s+/).filter((t) => t.length >= 3);
+  if (qTokens.length === 0) return true; // very short names: trust the scorer
+  const t = norm(title);
+  const tTokens = new Set(t.split(/\s+/));
+  for (const qt of qTokens) {
+    if (tTokens.has(qt)) return true;
+    if (t.includes(qt)) return true;
+    for (const tt of tTokens) if (tt.length >= 3 && qt.includes(tt)) return true;
+  }
+  return false;
+}
+
+/**
+ * Decide the final style + size for a node. When `icon` resolves, use the icon
+ * style (overlaid with a dark-preset font fix); otherwise a GENERIC shape by
+ * `kind` carrying the label. `resolved` reports whether an icon was found (used
+ * by the acceptance test that asserts no empty squares).
+ */
+export function resolveNodeStyle(
+  preset: PresetData,
+  node: GraphNode,
+): { style: string; w: number; h: number; iconResolved: boolean } {
+  if (node.icon) {
+    const shape = resolveIcon(node.icon);
+    if (shape) {
+      return {
+        style: iconNodeStyle(preset, shape.style),
+        w: shape.w,
+        h: shape.h,
+        iconResolved: true,
+      };
+    }
+  }
+  // Generic shape by kind, sized to the label so a long label never overflows.
+  const w = Math.max(120, estimateLabelWidth(node.label) + 32);
+  return { style: genericNodeStyle(preset, node.kind), w, h: 60, iconResolved: false };
+}
+
+/** Rough rendered width of the longest label line at 12px (~0.6em/glyph). */
+function estimateLabelWidth(label: string): number {
+  const lines = label.split(/\r?\n|&#xa;|<br\s*\/?>/i);
+  let longest = 0;
+  for (const l of lines) longest = Math.max(longest, l.trim().length);
+  return Math.ceil(longest * 12 * 0.6);
+}
+
+// --- graph -> ELK-JSON -----------------------------------------------------
+
+interface ElkNode {
+  id: string;
+  width?: number;
+  height?: number;
+  x?: number;
+  y?: number;
+  children?: ElkNode[];
+  layoutOptions?: Record<string, string>;
+}
+interface ElkEdge {
+  id: string;
+  sources: string[];
+  targets: string[];
+}
+interface ElkGraph extends ElkNode {
+  edges?: ElkEdge[];
+}
+
+const ELK_DIRECTION: Record<string, string> = {
+  LR: "RIGHT",
+  RL: "LEFT",
+  TB: "DOWN",
+  BT: "UP",
+};
+
+/** Sizes resolved per node id (from resolveNodeStyle), fed to the ELK mapper. */
+export interface NodeSize {
+  w: number;
+  h: number;
+}
+
+/**
+ * Build the ELK graph from the semantic graph + resolved node sizes. Compound
+ * group nodes nest their members (a group may itself nest in another group).
+ * `only` restricts the graph to a subset of node ids (used by the incremental
+ * path to lay out ONLY the new nodes). Layout HINTS (`layer`/`sameLayerAs`/
+ * `pinned`) are NOT encoded as ELK constraints here — ELK's constraint knobs are
+ * unreliable across versions — they are enforced deterministically AFTER layout
+ * by applyHints, which is exact and testable.
+ */
+export function graphToElk(
+  graph: Graph,
+  sizes: Map<string, NodeSize>,
+  opts: { only?: Set<string> } = {},
+): ElkGraph {
+  const direction = ELK_DIRECTION[graph.direction ?? "LR"] ?? "RIGHT";
+  const only = opts.only;
+  const include = (id: string) => !only || only.has(id);
+
+  const makeNode = (n: GraphNode): ElkNode => {
+    const size = sizes.get(n.id) ?? { w: 140, h: 60 };
+    return { id: n.id, width: size.w, height: size.h };
+  };
+
+  // Group children nest under their group node; ungrouped nodes are roots.
+  const groupNode = new Map<string, ElkNode>();
+  const usedGroups = new Set<string>();
+  for (const g of graph.groups ?? []) {
+    const size = sizes.get(g.id) ?? { w: 200, h: 150 };
+    groupNode.set(g.id, {
+      id: g.id,
+      width: size.w,
+      height: size.h,
+      children: [],
+      layoutOptions: {
+        "elk.algorithm": "layered",
+        "elk.direction": direction,
+        "elk.padding": "[top=40,left=30,bottom=30,right=30]",
+        "elk.layered.spacing.nodeNodeBetweenLayers": "170",
+        "elk.spacing.nodeNode": "170",
+      },
+    });
+  }
+  const roots: ElkNode[] = [];
+  for (const n of graph.nodes) {
+    if (!include(n.id)) continue;
+    const en = makeNode(n);
+    if (n.group && groupNode.has(n.group)) {
+      groupNode.get(n.group)!.children!.push(en);
+      usedGroups.add(n.group);
+    } else {
+      roots.push(en);
+    }
+  }
+  // Nest group nodes into their parent group (a subnet inside a VPC); groups
+  // with no parent group become roots. Only groups that hold an included node.
+  const groupIdSet = new Set((graph.groups ?? []).map((g) => g.id));
+  for (const g of graph.groups ?? []) {
+    if (only && !usedGroups.has(g.id)) continue;
+    const en = groupNode.get(g.id)!;
+    if (g.group && groupIdSet.has(g.group) && g.group !== g.id && (!only || usedGroups.has(g.group))) {
+      groupNode.get(g.group)!.children!.push(en);
+    } else {
+      roots.push(en);
+    }
+  }
+
+  // Edges: endpoints may be nodes or groups; INCLUDE_CHILDREN spans the nesting.
+  const validIds = new Set<string>([
+    ...graph.nodes.filter((n) => include(n.id)).map((n) => n.id),
+    ...(graph.groups ?? []).map((g) => g.id),
+  ]);
+  const edges: ElkEdge[] = [];
+  (graph.edges ?? []).forEach((e, i) => {
+    if (!validIds.has(e.from) || !validIds.has(e.to)) return;
+    edges.push({ id: `e${i}`, sources: [e.from], targets: [e.to] });
+  });
+
+  const rootOptions: Record<string, string> = {
+    "elk.algorithm": "layered",
+    "elk.direction": direction,
+    "elk.hierarchyHandling": "INCLUDE_CHILDREN",
+    "elk.layered.spacing.nodeNodeBetweenLayers": "170",
+    "elk.spacing.nodeNode": "170",
+    "elk.spacing.edgeNode": "40",
+    "elk.spacing.edgeEdge": "30",
+    "elk.padding": "[top=20,left=20,bottom=20,right=20]",
+  };
+
+  return { id: "root", layoutOptions: rootOptions, children: roots, edges };
+}
+
+/**
+ * Enforce the layout hints DETERMINISTICALLY on ELK's output (mutates `geo`):
+ *   - `sameLayerAs`: snap the dependent node's LAYER-AXIS coordinate to its
+ *     anchor's, so the pair lands in the same layer (x for LR/RL, y for TB/BT).
+ *     `layer` groups nodes with the same index onto the same anchor coordinate.
+ *   - `pinned`: override the node's coordinate with the exact pinned {x,y}.
+ * Applied only to top-level (ungrouped) nodes, whose ELK coords are absolute.
+ */
+export function applyHints(
+  graph: Graph,
+  geo: Map<string, { x: number; y: number; w: number; h: number }>,
+): void {
+  const dir = graph.direction ?? "LR";
+  const layerAxis: "x" | "y" = dir === "TB" || dir === "BT" ? "y" : "x";
+  // The perpendicular (cross-layer) axis: members snapped onto one layer must be
+  // spread along THIS axis so they don't stack onto the same point.
+  const crossAxis: "x" | "y" = layerAxis === "x" ? "y" : "x";
+  const crossSize: "w" | "h" = crossAxis === "x" ? "w" : "h";
+  const grouped = new Set(
+    graph.nodes.filter((n) => n.group).map((n) => n.id),
+  );
+
+  // sameLayerAs / layer: co-assign the layer-axis coordinate.
+  // Build the effective layer key per node, then pick a representative coord.
+  const layerKeyOf = new Map<string, string>();
+  const explicitLayer = new Map<string, number>();
+  for (const n of graph.nodes) if (n.layer != null) explicitLayer.set(n.id, n.layer);
+  const byId = new Map(graph.nodes.map((n) => [n.id, n]));
+  const resolveKey = (n: GraphNode): string | null => {
+    if (explicitLayer.has(n.id)) return `L${explicitLayer.get(n.id)}`;
+    const seen = new Set<string>([n.id]);
+    let cur: GraphNode | undefined = n;
+    while (cur && cur.sameLayerAs != null && !seen.has(cur.sameLayerAs)) {
+      seen.add(cur.sameLayerAs);
+      const t = byId.get(cur.sameLayerAs);
+      if (!t) break;
+      if (explicitLayer.has(t.id)) return `L${explicitLayer.get(t.id)}`;
+      cur = t;
+    }
+    // A sameLayerAs chain with no explicit layer: key on the chain's root id.
+    if (n.sameLayerAs != null) {
+      let root = n.id;
+      const s2 = new Set<string>([n.id]);
+      let c: GraphNode | undefined = n;
+      while (c && c.sameLayerAs != null && !s2.has(c.sameLayerAs)) {
+        s2.add(c.sameLayerAs);
+        root = c.sameLayerAs;
+        c = byId.get(c.sameLayerAs);
+      }
+      return `C${root}`;
+    }
+    return null;
+  };
+  for (const n of graph.nodes) {
+    if (grouped.has(n.id)) continue; // group children are relative — skip
+    const key = resolveKey(n);
+    if (key) layerKeyOf.set(n.id, key);
+  }
+  // Group members of each layer key so we can snap AND spread them together.
+  const membersOf = new Map<string, string[]>();
+  for (const n of graph.nodes) {
+    const key = layerKeyOf.get(n.id);
+    if (key == null) continue;
+    if (!geo.has(n.id)) continue;
+    (membersOf.get(key) ?? membersOf.set(key, []).get(key)!).push(n.id);
+  }
+  // For each layer key: snap every member to the FIRST member's layer-axis coord,
+  // then SPREAD them along the perpendicular (cross-layer) axis with a >=
+  // MIN_SHAPE_GAP gap. Without the spread, a sameLayerAs chain whose nodes ELK
+  // happened to give the same cross-axis coordinate would collapse onto one point
+  // -> shape-overlap + edge-through-shape quality warnings (breaking the
+  // "0 warnings by construction" guarantee for these AUTO-positioned hints). We
+  // start from the members' minimum cross-axis coord and stack them with a gap
+  // of MIN_SHAPE_GAP beyond each shape's cross-axis size.
+  for (const [key, members] of membersOf) {
+    if (members.length === 0) continue;
+    // Snap layer-axis coord to the first member.
+    const repCoord = geo.get(members[0])![layerAxis];
+    // Preserve the members' existing relative order along the cross axis so the
+    // spread stays visually stable, then re-lay them contiguously.
+    const sorted = [...members].sort(
+      (a, b) => geo.get(a)![crossAxis] - geo.get(b)![crossAxis],
+    );
+    let cursor = geo.get(sorted[0])![crossAxis];
+    for (const id of sorted) {
+      const g = geo.get(id)!;
+      g[layerAxis] = repCoord;
+      g[crossAxis] = cursor;
+      cursor += g[crossSize] + MIN_SHAPE_GAP;
+    }
+    void key;
+  }
+
+  // pinned: exact override (wins over any layer snap). Explicit user coordinates
+  // are user intent, but CLAMP to non-negative so an out-of-bounds pin (e.g.
+  // x:-500) never renders off-canvas. Two user-pinned nodes at the same point is
+  // user error the server can't silently relocate — the assembler docstring
+  // documents that explicit pins are user-directed and MAY warn (see #423/#425
+  // acceptance: the "0 quality-warnings by construction" guarantee is for
+  // AUTO-LAYOUT, not for coordinates the user pinned by hand).
+  for (const n of graph.nodes) {
+    if (!n.pinned) continue;
+    const px = Math.max(0, n.pinned.x);
+    const py = Math.max(0, n.pinned.y);
+    const g = geo.get(n.id);
+    if (g) {
+      g.x = px;
+      g.y = py;
+    } else {
+      const sz = { w: 140, h: 60 };
+      geo.set(n.id, { x: px, y: py, w: sz.w, h: sz.h });
+    }
+  }
+}
+
+/**
+ * Incremental variant of applyHints: apply `pinned` only to NEW nodes (those
+ * absent from `existing`); an existing node's coordinates are NEVER changed
+ * (acceptance #3). sameLayerAs/layer snapping is intentionally skipped in the
+ * incremental path — moving a new node's layer axis could still be desired, but
+ * it must never move an existing cell, so we keep the incremental contract
+ * simple: existing cells are frozen, new pinned nodes honour their pin.
+ */
+export function applyHintsForNew(
+  graph: Graph,
+  geo: Map<string, { x: number; y: number; w: number; h: number }>,
+  existing: Map<string, { x: number; y: number }>,
+): void {
+  for (const n of graph.nodes) {
+    if (existing.has(n.id)) continue; // never move an existing cell
+    if (!n.pinned) continue;
+    const px = Math.max(0, n.pinned.x); // clamp out-of-bounds pins non-negative
+    const py = Math.max(0, n.pinned.y);
+    const g = geo.get(n.id);
+    if (g) {
+      g.x = px;
+      g.y = py;
+    } else {
+      geo.set(n.id, { x: px, y: py, w: 140, h: 60 });
+    }
+  }
+}
+
+// --- layout runner ---------------------------------------------------------
+
+const ELK_TIMEOUT_MS = 5000;
+
+/**
+ * Run ELK over the mapped graph and return computed geometry per id (coords are
+ * parent-relative, matching mxGraph's convention for container children). On any
+ * ELK failure/timeout the returned map is empty and the caller falls back to a
+ * deterministic grid placement (so the write never fails on a layout hiccup).
+ */
+export async function runElk(
+  elk: ElkGraph,
+): Promise<Map<string, { x: number; y: number; w: number; h: number }>> {
+  const geo = new Map<string, { x: number; y: number; w: number; h: number }>();
+  let timer: ReturnType<typeof setTimeout> | undefined;
+  try {
+    const Ctor: any = (ELK as any).default ?? ELK;
+    const inst = new Ctor();
+    const timeout = new Promise<never>((_, reject) => {
+      timer = setTimeout(() => reject(new Error("ELK timed out")), ELK_TIMEOUT_MS);
+    });
+    const laid = (await Promise.race([inst.layout(elk as any), timeout])) as ElkGraph;
+    const walk = (n: ElkNode) => {
+      if (n.id !== "root") {
+        geo.set(n.id, {
+          x: Math.round(n.x ?? 0),
+          y: Math.round(n.y ?? 0),
+          w: Math.round(n.width ?? 140),
+          h: Math.round(n.height ?? 60),
+        });
+      }
+      for (const c of n.children ?? []) walk(c);
+    };
+    walk(laid);
+  } catch {
+    return new Map(); // best-effort: empty -> caller uses fallback grid.
+  } finally {
+    if (timer) clearTimeout(timer);
+  }
+  return geo;
+}
+
+// --- XML assembler ---------------------------------------------------------
+
+/** Order groups so a parent group always precedes its nested children. */
+function topoSortGroups(groups: GraphGroup[], groupIds: Set<string>): GraphGroup[] {
+  const byId = new Map(groups.map((g) => [g.id, g]));
+  const out: GraphGroup[] = [];
+  const done = new Set<string>();
+  const visit = (g: GraphGroup, stack: Set<string>) => {
+    if (done.has(g.id)) return;
+    if (stack.has(g.id)) return; // cycle guard
+    stack.add(g.id);
+    if (g.group && groupIds.has(g.group) && g.group !== g.id) {
+      const parent = byId.get(g.group);
+      if (parent) visit(parent, stack);
+    }
+    stack.delete(g.id);
+    if (!done.has(g.id)) {
+      done.add(g.id);
+      out.push(g);
+    }
+  };
+  for (const g of groups) visit(g, new Set());
+  return out;
+}
+
+function xmlEscapeAttr(s: string): string {
+  return s
+    .replace(/&/g, "&amp;")
+    .replace(/</g, "&lt;")
+    .replace(/>/g, "&gt;")
+    .replace(/"/g, "&quot;")
+    .replace(/\r\n|\r|\n/g, "&#xa;"); // literal newline -> the linter-approved entity
+}
+
+export interface AssembleResult {
+  modelXml: string;
+  iconsResolved: number;
+  iconsMissing: string[];
+}
+
+/**
+ * Assemble the final mxGraphModel XML from the graph + resolved styles/coords.
+ * Guarantees BY CONSTRUCTION that the #423 linter passes:
+ *   - id=0 and id=1(parent=0) sentinels;
+ *   - each node/group is vertex="1" (containers get container=1 via the style);
+ *   - each edge is edge="1" with a child <mxGeometry relative="1" as="geometry"/>;
+ *   - group children set parent=<groupId> and RELATIVE coords; an edge between
+ *     two different parents is parent="1";
+ *   - labels are XML-escaped and any newline is &#xa;.
+ * `geo` may be empty (ELK failed) — then a deterministic grid is used so the
+ * output is still valid and non-overlapping (>=170px stride).
+ *
+ * QUALITY-WARNING GUARANTEE: the "0 quality-warnings by construction" promise
+ * holds for AUTO-LAYOUT — ELK spacing plus applyHints' cross-axis spread for the
+ * server-positioned `layer`/`sameLayerAs` hints keep shapes >=MIN_SHAPE_GAP
+ * apart. It does NOT extend to explicit `pinned` coordinates: those are
+ * user-directed, so two nodes the user pins to the same/overlapping point are
+ * user error the server honours verbatim (only clamped non-negative) and MAY
+ * therefore produce a quality warning.
+ */
+export function assembleModel(
+  graph: Graph,
+  opts: {
+    preset: PresetData;
+    styles: Map<string, { style: string; w: number; h: number; iconResolved: boolean }>;
+    geo: Map<string, { x: number; y: number; w: number; h: number }>;
+  },
+): AssembleResult {
+  const { preset, styles, geo } = opts;
+  const groupIds = new Set((graph.groups ?? []).map((g) => g.id));
+  const nodeById = new Map(graph.nodes.map((n) => [n.id, n]));
+
+  // Fallback grid when ELK produced nothing: lay ungrouped nodes on a grid with
+  // a 190px stride (>150 gap). Grouped nodes/groups are placed inside their group.
+  const fallback = geo.size === 0;
+  const gridPos = (i: number) => ({ x: 40 + (i % 5) * 200, y: 40 + Math.floor(i / 5) * 140 });
+
+  const cells: string[] = ['<mxCell id="0"/>', '<mxCell id="1" parent="0"/>'];
+
+  // Groups first (they are parents of their members). A nested group sets
+  // parent=<parentGroupId>; emit parents before children so parent-exists holds.
+  let gi = 0;
+  const groupGeo = new Map<string, { x: number; y: number; w: number; h: number }>();
+  const orderedGroups = topoSortGroups(graph.groups ?? [], groupIds);
+  for (const g of orderedGroups) {
+    const gg = geo.get(g.id) ?? { ...gridPos(gi++), w: 320, h: 220 };
+    groupGeo.set(g.id, gg);
+    const style = groupStyle(preset);
+    const gParent = g.group && groupIds.has(g.group) && g.group !== g.id ? g.group : "1";
+    cells.push(
+      `<mxCell id="${xmlEscapeAttr(g.id)}" value="${xmlEscapeAttr(g.label)}" style="${style}" vertex="1" parent="${xmlEscapeAttr(gParent)}">` +
+        `<mxGeometry x="${gg.x}" y="${gg.y}" width="${gg.w}" height="${gg.h}" as="geometry"/></mxCell>`,
+    );
+  }
+
+  // Nodes. A grouped node's coords are RELATIVE to its group (ELK already
+  // returns child coords relative to the parent; for the fallback grid we place
+  // children on a small in-group grid).
+  let ungrouped = (graph.groups?.length ?? 0);
+  const inGroupIndex = new Map<string, number>();
+  let iconsResolved = 0;
+  const iconsMissing: string[] = [];
+  for (const n of graph.nodes) {
+    const st = styles.get(n.id)!;
+    if (n.icon) {
+      if (st.iconResolved) iconsResolved++;
+      else iconsMissing.push(n.id);
+    }
+    let x: number;
+    let y: number;
+    const g = geo.get(n.id);
+    if (g && !fallback) {
+      x = g.x;
+      y = g.y;
+    } else if (n.group && groupIds.has(n.group)) {
+      const k = inGroupIndex.get(n.group) ?? 0;
+      inGroupIndex.set(n.group, k + 1);
+      x = 30 + (k % 3) * 180;
+      y = 40 + Math.floor(k / 3) * 120;
+    } else {
+      const p = gridPos(ungrouped++);
+      x = p.x;
+      y = p.y;
+    }
+    const parent = n.group && groupIds.has(n.group) ? n.group : "1";
+    cells.push(
+      `<mxCell id="${xmlEscapeAttr(n.id)}" value="${xmlEscapeAttr(n.label)}" style="${st.style}" vertex="1" parent="${xmlEscapeAttr(parent)}">` +
+        `<mxGeometry x="${x}" y="${y}" width="${st.w}" height="${st.h}" as="geometry"/></mxCell>`,
+    );
+  }
+
+  // Edges. parent="1" whenever the two endpoints have different container
+  // parents (or either is a group); otherwise the shared group id.
+  (graph.edges ?? []).forEach((e, i) => {
+    const style = edgeStyle(preset, e.kind);
+    const fromNode = nodeById.get(e.from);
+    const toNode = nodeById.get(e.to);
+    const fromParent = fromNode?.group && groupIds.has(fromNode.group) ? fromNode.group : "1";
+    const toParent = toNode?.group && groupIds.has(toNode.group) ? toNode.group : "1";
+    const parent = fromParent === toParent ? fromParent : "1";
+    const label = e.label ? ` value="${xmlEscapeAttr(e.label)}"` : "";
+    cells.push(
+      `<mxCell id="ge${i}"${label} style="${style}" edge="1" parent="${xmlEscapeAttr(parent)}" ` +
+        `source="${xmlEscapeAttr(e.from)}" target="${xmlEscapeAttr(e.to)}">` +
+        `<mxGeometry relative="1" as="geometry"/></mxCell>`,
+    );
+  });
+
+  const modelAttrs =
+    'dx="0" dy="0" grid="1" gridSize="10" page="1" pageWidth="850" pageHeight="1100" adaptiveColors="auto"';
+  const modelXml = `<mxGraphModel ${modelAttrs}><root>${cells.join("")}</root></mxGraphModel>`;
+  return { modelXml, iconsResolved, iconsMissing };
+}
+
+// --- incremental merge -----------------------------------------------------
+
+let _mergeWindow: any = null;
+function mergeWindow(): any {
+  if (!_mergeWindow) _mergeWindow = new JSDOM("").window;
+  return _mergeWindow;
+}
+
+/**
+ * Merge the freshly-assembled graph XML with the EXISTING diagram model so an
+ * incremental "add a node" call never drops a hand-placed cell. `assembleModel`
+ * emits ONLY the passed graph's cells; on its own it would replace the whole
+ * model, wiping any existing cell the caller didn't re-list. This splices every
+ * existing cell that the graph does NOT re-list (preserved verbatim: coords,
+ * style, edges) into the assembled root:
+ *   - id in the graph  -> the graph's (re-laid) cell wins (already assembled;
+ *     coords are frozen for existing ids via the incremental geo path);
+ *   - id NOT in the graph -> the existing cell is preserved verbatim;
+ *   - a graph node absent from the existing model -> added (offset clear).
+ * The sentinels ("0"/"1") come from the assembled model and are never doubled.
+ */
+function mergeExistingCells(
+  assembledXml: string,
+  existingModelXml: string,
+  graph: Graph,
+): string {
+  const win = mergeWindow();
+  const parser = new win.DOMParser();
+  const existingDoc = parser.parseFromString(existingModelXml, "application/xml");
+  if (existingDoc.getElementsByTagName("parsererror").length > 0) {
+    // Existing model unreadable: fall back to the assembled model alone (still a
+    // valid diagram — better than throwing on a corrupt prior file).
+    return assembledXml;
+  }
+  const assembledDoc = parser.parseFromString(assembledXml, "application/xml");
+  const root = assembledDoc.getElementsByTagName("root")[0];
+  if (!root) return assembledXml;
+
+  // Ids the assembled model already emitted (graph nodes/groups/edges + sentinels).
+  const assembledIds = new Set<string>();
+  for (const el of Array.from(root.getElementsByTagName("mxCell")) as any[]) {
+    const id = el.getAttribute("id");
+    if (id) assembledIds.add(id);
+  }
+  // The graph's own ids: any existing cell with one of these is superseded by the
+  // assembled version and must NOT be re-imported.
+  const graphIds = new Set<string>([
+    ...graph.nodes.map((n) => n.id),
+    ...(graph.groups ?? []).map((g) => g.id),
+  ]);
+
+  const existingCells = Array.from(
+    existingDoc.getElementsByTagName("mxCell"),
+  ) as any[];
+  for (const el of existingCells) {
+    const id = el.getAttribute("id") ?? "";
+    if (id === "0" || id === "1") continue; // sentinels come from the assembled model
+    if (graphIds.has(id)) continue; // graph re-lists it -> assembled version wins
+    if (assembledIds.has(id)) continue; // id collision guard -> keep assembled
+    root.appendChild(assembledDoc.importNode(el, true));
+    assembledIds.add(id);
+  }
+
+  const ser = new win.XMLSerializer();
+  return ser.serializeToString(assembledDoc.documentElement);
+}
+
+// --- top-level: graph -> mxGraphModel XML ----------------------------------
+
+export interface BuildFromGraphResult {
+  modelXml: string;
+  iconsResolved: number;
+  iconsMissing: string[];
+  layout: LayoutMode;
+}
+
+/**
+ * The full server-side pipeline: validate -> resolve styles/icons -> map to ELK
+ * -> run ELK (or fall back) -> assemble linter-clean XML. `existingCoords` is
+ * supplied for `layout:"incremental"` (the coordinates of the diagram's current
+ * cells, so they are preserved and only new nodes are placed). `existingModelXml`
+ * is the current diagram's full model XML — in incremental mode every existing
+ * cell the graph does NOT re-list is MERGED back in verbatim so a hand-placed
+ * cell is never dropped (WARNING #4). Pure — no network.
+ */
+export async function buildFromGraph(
+  graph: Graph,
+  layout: LayoutMode = "full",
+  existingCoords?: Map<string, { x: number; y: number }>,
+  existingModelXml?: string,
+): Promise<BuildFromGraphResult> {
+  validateGraph(graph);
+  const preset = getPreset(graph.preset);
+
+  // Resolve every node's style + size (icon or generic-by-kind).
+  const styles = new Map<
+    string,
+    { style: string; w: number; h: number; iconResolved: boolean }
+  >();
+  const sizes = new Map<string, NodeSize>();
+  for (const n of graph.nodes) {
+    const s = resolveNodeStyle(preset, n);
+    styles.set(n.id, s);
+    sizes.set(n.id, { w: s.w, h: s.h });
+  }
+  // Group sizes: seed a min box; ELK computes the real size when it lays out.
+  for (const g of graph.groups ?? []) sizes.set(g.id, { w: 240, h: 180 });
+
+  let geo = new Map<string, { x: number; y: number; w: number; h: number }>();
+
+  if (layout === "incremental" && existingCoords && existingCoords.size > 0) {
+    // INCREMENTAL: keep every existing cell's coords VERBATIM (acceptance #3 —
+    // never move a hand-arranged cell) and lay out ONLY the new nodes, then
+    // offset that block clear of the existing bbox so nothing overlaps.
+    for (const [id, c] of existingCoords) {
+      const sz = sizes.get(id) ?? { w: 140, h: 60 };
+      geo.set(id, { x: c.x, y: c.y, w: sz.w, h: sz.h });
+    }
+    const newIds = new Set(
+      graph.nodes.filter((n) => !existingCoords.has(n.id)).map((n) => n.id),
+    );
+    if (newIds.size > 0) {
+      const elk = graphToElk(graph, sizes, { only: newIds });
+      const laid = await runElk(elk);
+      // Place the new block below the existing content (a clear >=170px gap).
+      let maxY = 0;
+      for (const c of existingCoords.values()) maxY = Math.max(maxY, c.y);
+      const offsetY = maxY + 200;
+      for (const [id, g] of laid) {
+        if (newIds.has(id)) geo.set(id, { ...g, y: g.y + offsetY });
+        else if (!geo.has(id)) geo.set(id, g); // a new group container
+      }
+    }
+    // Hints still apply to NEW pinned nodes only (existing ones stay put).
+    applyHintsForNew(graph, geo, existingCoords);
+  } else if (layout !== "none") {
+    const elk = graphToElk(graph, sizes);
+    geo = await runElk(elk);
+    applyHints(graph, geo);
+  } else if (existingCoords) {
+    // layout:"none" with prior coords -> keep them verbatim.
+    for (const [id, c] of existingCoords) {
+      const sz = sizes.get(id) ?? { w: 140, h: 60 };
+      geo.set(id, { x: c.x, y: c.y, w: sz.w, h: sz.h });
+    }
+  }
+
+  const assembled = assembleModel(graph, { preset, styles, geo });
+  // In incremental mode, splice back every existing cell the graph didn't
+  // re-list so an "add one node" call preserves the user's manual layout.
+  let modelXml = assembled.modelXml;
+  if (
+    layout === "incremental" &&
+    existingModelXml &&
+    existingCoords &&
+    existingCoords.size > 0
+  ) {
+    modelXml = mergeExistingCells(modelXml, existingModelXml, graph);
+  }
+  return {
+    modelXml,
+    iconsResolved: assembled.iconsResolved,
+    iconsMissing: assembled.iconsMissing,
+    layout,
+  };
+}

packages/mcp/src/lib/drawio-mermaid.ts

@@ -0,0 +1,347 @@
+// Pure Mermaid `flowchart` -> graph-JSON parser for `drawioFromMermaid` (issue
+// #425, stage 3, OPTIONAL). The escape clause in the issue: convert WITHOUT
+// Electron/draw.io-CLI, so a pure text parser only. It handles the common wiki
+// flowchart subset — node shapes, labelled/dashed edges, subgraphs (-> groups),
+// and the direction header — and emits a Graph the drawioFromGraph pipeline
+// renders as an EDITABLE draw.io diagram. Anything beyond flowchart (sequence /
+// class / state) throws a clear error so the model falls back to drawioFromGraph.
+//
+// DELIBERATELY NARROW: this is not a full Mermaid grammar (Mermaid's own parser
+// is a 100KB+ browser dependency). It covers `flowchart`/`graph` with the node
+// shapes and edge arrows that show up in practice; unusual syntax is skipped
+// rather than mis-parsed, and a diagram that yields no nodes throws.
+
+import type { Graph, GraphNode, GraphEdge, GraphGroup } from "./drawio-graph.js";
+
+export class MermaidParseError extends Error {
+  constructor(message: string) {
+    super(`drawioFromMermaid: ${message}`);
+    this.name = "MermaidParseError";
+  }
+}
+
+// Input-size bounds applied BEFORE parsing. Without them a pathological mermaid
+// string (e.g. 300000 connection lines, or 20000 nested `subgraph`s) builds a
+// huge intermediate node/edge/group structure that OOM-crashes the worker — the
+// downstream validateGraph caps in drawio-graph can't help because the parser
+// exhausts the heap constructing the intermediate FIRST. These caps reject the
+// over-limit input fast, before a single line is parsed.
+const MAX_MERMAID_CHARS = 200_000; // ~200 KB of source is far beyond any real diagram.
+const MAX_MERMAID_LINES = 20_000;
+const MAX_MERMAID_GROUPS = 500; // parity with drawio-graph's MAX_GRAPH_GROUPS.
+// Per connection line, the number of chained nodes we will expand (`A-->B-->C`).
+const MAX_CHAIN_NODES = 500;
+
+const DIRECTIONS: Record<string, Graph["direction"]> = {
+  LR: "LR",
+  RL: "RL",
+  TB: "TB",
+  TD: "TB",
+  BT: "BT",
+};
+
+/**
+ * Node-shape delimiters -> a semantic `kind`. Mermaid encodes shape in the
+ * bracket style; we map the common ones to the palette kinds so the diagram is
+ * colored meaningfully (a decision/diamond -> queue, a database cylinder -> db,
+ * a rounded/stadium -> service, a subroutine/hexagon -> gateway, default rect ->
+ * service). The label text lives between the delimiters.
+ */
+interface ShapeDef {
+  open: string;
+  close: string;
+  kind: string;
+}
+// Order matters: longer/multi-char delimiters first so "([" beats "(".
+const SHAPES: ShapeDef[] = [
+  { open: "([", close: "])", kind: "service" }, // stadium
+  { open: "[[", close: "]]", kind: "gateway" }, // subroutine
+  { open: "[(", close: ")]", kind: "db" }, // cylinder-ish / database
+  { open: "((", close: "))", kind: "external" }, // circle
+  { open: "{{", close: "}}", kind: "gateway" }, // hexagon
+  { open: "[", close: "]", kind: "service" }, // rectangle
+  { open: "(", close: ")", kind: "service" }, // rounded
+  { open: "{", close: "}", kind: "queue" }, // rhombus / decision
+  { open: ">", close: "]", kind: "external" }, // asymmetric flag
+];
+
+/** Strip Mermaid label quoting/escapes and normalise whitespace. */
+function cleanLabel(raw: string): string {
+  let s = raw.trim();
+  if (
+    (s.startsWith('"') && s.endsWith('"')) ||
+    (s.startsWith("'") && s.endsWith("'"))
+  ) {
+    s = s.slice(1, -1);
+  }
+  return s.replace(/<br\s*\/?>/gi, " ").replace(/\s+/g, " ").trim();
+}
+
+/** A single edge-arrow spec: its regex and the resulting edge `kind`. */
+interface ArrowDef {
+  re: RegExp;
+  kind: string;
+}
+// Dotted arrows (`-.->`) -> async; thick (`==>`) stay sync; normal `-->`/`---`.
+// Each captures an optional `|label|` OR inline label between the two arrow
+// halves. Applied to the segment between two node tokens.
+const ARROWS: ArrowDef[] = [
+  { re: /-\.->|-\.-/, kind: "async" },
+  { re: /==>|===/, kind: "sync" },
+  { re: /-->|---/, kind: "sync" },
+];
+
+interface ParsedRef {
+  id: string;
+  node?: GraphNode;
+}
+
+/**
+ * Parse a single node token like `A`, `A[Label]`, `db[(Orders)]`, `d{Choose}`.
+ * Returns the id and, when the token declares a shape/label, a GraphNode.
+ */
+function parseNodeToken(token: string): ParsedRef | null {
+  const t = token.trim();
+  if (t === "") return null;
+  for (const shape of SHAPES) {
+    const oi = t.indexOf(shape.open);
+    if (oi <= 0) continue;
+    if (!t.endsWith(shape.close)) continue;
+    const id = t.slice(0, oi).trim();
+    const label = cleanLabel(t.slice(oi + shape.open.length, t.length - shape.close.length));
+    if (!id) return null;
+    return { id, node: { id, label: label || id, kind: shape.kind } };
+  }
+  // Bare id (no shape declared here — may be defined elsewhere).
+  if (/^[A-Za-z0-9_.-]+$/.test(t)) return { id: t };
+  return null;
+}
+
+/**
+ * Split a connection line into [leftToken, arrowSegment, rightToken]. Returns
+ * null if the line has no arrow. The arrow segment may embed a label as
+ * `-->|text|` or `-- text -->`.
+ */
+function splitConnection(
+  line: string,
+): { left: string; right: string; kind: string; label?: string } | null {
+  for (const arrow of ARROWS) {
+    // Find the arrow occurrence. Support a mid-arrow label: `A -- text --> B`.
+    const m = arrow.re.exec(line);
+    if (!m) continue;
+    const idx = m.index;
+    let left = line.slice(0, idx).trim();
+    let rest = line.slice(idx + m[0].length).trim();
+    let label: string | undefined;
+    // Pipe label: `-->|HTTPS| B`.
+    const pipe = /^\|([^|]*)\|\s*(.*)$/.exec(rest);
+    if (pipe) {
+      label = cleanLabel(pipe[1]);
+      rest = pipe[2].trim();
+    }
+    // Mid-arrow label on the left side: `A -- text` before the arrow half.
+    const midLeft = /^(.*?)\s*--\s*(.+)$/.exec(left);
+    if (!label && midLeft && /-\.|--|==/.test(line.slice(0, idx))) {
+      // Only treat as a label when there's clearly text after `--`.
+      if (!/[\[\](){}]/.test(midLeft[2])) {
+        left = midLeft[1].trim();
+        label = cleanLabel(midLeft[2]);
+      }
+    }
+    if (!left || !rest) return null;
+    return { left, right: rest, kind: arrow.kind, label };
+  }
+  return null;
+}
+
+/**
+ * Parse Mermaid flowchart text into a Graph. Handles the header
+ * (`flowchart LR` / `graph TD`), `subgraph <id>[title] … end` blocks (-> groups),
+ * node declarations, and connection lines. Throws MermaidParseError for a
+ * non-flowchart diagram or when nothing parses.
+ */
+export function mermaidToGraph(mermaid: string): Graph {
+  if (typeof mermaid !== "string" || mermaid.trim() === "") {
+    throw new MermaidParseError("empty mermaid input");
+  }
+  // Size guards FIRST — bound the raw input before building any intermediate.
+  if (mermaid.length > MAX_MERMAID_CHARS) {
+    throw new MermaidParseError(
+      `input is ${mermaid.length} chars (max ${MAX_MERMAID_CHARS}); split the diagram or use drawioFromGraph`,
+    );
+  }
+  const rawLines = mermaid.split(/\r?\n/);
+  if (rawLines.length > MAX_MERMAID_LINES) {
+    throw new MermaidParseError(
+      `input has ${rawLines.length} lines (max ${MAX_MERMAID_LINES}); split the diagram or use drawioFromGraph`,
+    );
+  }
+  const nodes = new Map<string, GraphNode>();
+  const groups: GraphGroup[] = [];
+  const edges: GraphEdge[] = [];
+  let direction: Graph["direction"] = "LR";
+  let sawHeader = false;
+
+  // Stack of active subgraph ids (nesting); the top is the current group.
+  const groupStack: string[] = [];
+  let anonGroup = 0;
+
+  const ensureNode = (ref: ParsedRef) => {
+    const existing = nodes.get(ref.id);
+    if (ref.node) {
+      if (existing) {
+        // Fill in a label/kind if this token declared a shape and the prior didn't.
+        if (existing.label === existing.id && ref.node.label !== ref.node.id)
+          existing.label = ref.node.label;
+        if (!existing.kind) existing.kind = ref.node.kind;
+      } else {
+        nodes.set(ref.id, { ...ref.node });
+      }
+    } else if (!existing) {
+      nodes.set(ref.id, { id: ref.id, label: ref.id, kind: "service" });
+    }
+    // Assign to the current subgraph if inside one and not yet grouped.
+    const cur = groupStack[groupStack.length - 1];
+    const n = nodes.get(ref.id)!;
+    if (cur && n.group == null) n.group = cur;
+  };
+
+  for (const raw of rawLines) {
+    let line = raw.trim();
+    if (line === "" || line.startsWith("%%")) continue; // blank / comment
+
+    // Header.
+    const header = /^(flowchart|graph)\s+([A-Za-z]{2})\b/.exec(line);
+    if (header) {
+      sawHeader = true;
+      const dir = DIRECTIONS[header[2].toUpperCase()];
+      if (dir) direction = dir;
+      continue;
+    }
+    if (/^(sequenceDiagram|classDiagram|stateDiagram|erDiagram|gantt|pie|journey)\b/.test(line)) {
+      throw new MermaidParseError(
+        `only 'flowchart'/'graph' is supported (got '${line.split(/\s+/)[0]}'); use drawioFromGraph instead`,
+      );
+    }
+
+    // Subgraph open: `subgraph id [Title]` or `subgraph Title`.
+    const sg = /^subgraph\s+(.+)$/.exec(line);
+    if (sg) {
+      const spec = sg[1].trim();
+      let id: string;
+      let label: string;
+      const bracket = /^([A-Za-z0-9_.-]+)\s*\[(.+)\]$/.exec(spec);
+      if (bracket) {
+        id = bracket[1];
+        label = cleanLabel(bracket[2]);
+      } else if (/^[A-Za-z0-9_.-]+$/.test(spec)) {
+        id = spec;
+        label = spec;
+      } else {
+        id = `sg${anonGroup++}`;
+        label = cleanLabel(spec);
+      }
+      if (!groups.some((g) => g.id === id)) {
+        if (groups.length >= MAX_MERMAID_GROUPS) {
+          throw new MermaidParseError(
+            `too many subgraphs (max ${MAX_MERMAID_GROUPS}); use drawioFromGraph for a diagram this large`,
+          );
+        }
+        groups.push({ id, label, kind: "group" });
+      }
+      groupStack.push(id);
+      continue;
+    }
+    if (/^end\b/.test(line)) {
+      groupStack.pop();
+      continue;
+    }
+    // `direction LR` inside a subgraph — apply to the top-level direction.
+    const innerDir = /^direction\s+([A-Za-z]{2})\b/.exec(line);
+    if (innerDir) {
+      const dir = DIRECTIONS[innerDir[1].toUpperCase()];
+      if (dir) direction = dir;
+      continue;
+    }
+    // Style/class/click directives: ignore (no visual mapping in our palette).
+    if (/^(style|classDef|class|click|linkStyle)\b/.test(line)) continue;
+
+    // Strip a trailing semicolon.
+    if (line.endsWith(";")) line = line.slice(0, -1).trim();
+
+    // Connection line (possibly chained: A --> B --> C).
+    const conn = splitConnection(line);
+    if (conn) {
+      // Handle a simple chain by re-splitting the right side.
+      let leftTok = conn.left;
+      let seg: typeof conn | null = conn;
+      let guard = 0;
+      while (seg) {
+        if (guard++ >= MAX_CHAIN_NODES) {
+          // Don't silently drop the tail of an over-long chain — surface it so
+          // the model knows the diagram was too large rather than getting a
+          // quietly-truncated result.
+          throw new MermaidParseError(
+            `a single connection chain exceeds ${MAX_CHAIN_NODES} nodes; split it or use drawioFromGraph`,
+          );
+        }
+        const leftRef = parseNodeToken(leftTok);
+        // The right side may itself contain another arrow (a chain).
+        const nextSeg = splitConnection(seg.right);
+        const rightTokenStr = nextSeg ? seg.right.slice(0, splitIndex(seg.right)) : seg.right;
+        const rightRef = parseNodeToken(nextSeg ? nextSeg.left : seg.right);
+        if (leftRef && rightRef) {
+          ensureNode(leftRef);
+          ensureNode(rightRef);
+          edges.push({
+            from: leftRef.id,
+            to: rightRef.id,
+            label: seg.label,
+            kind: seg.kind,
+          });
+        }
+        if (!nextSeg) break;
+        leftTok = nextSeg.left;
+        seg = nextSeg;
+        void rightTokenStr;
+      }
+      continue;
+    }
+
+    // Standalone node declaration `A[Label]` OR a bare member ref `C` inside a
+    // subgraph (which claims that node for the current group).
+    const nodeRef = parseNodeToken(line);
+    if (nodeRef && (nodeRef.node || groupStack.length > 0)) {
+      ensureNode(nodeRef);
+      continue;
+    }
+    // Unknown line: skip silently (robustness over strictness).
+  }
+
+  if (!sawHeader && nodes.size === 0) {
+    throw new MermaidParseError(
+      "input does not look like a mermaid flowchart (no 'flowchart'/'graph' header and no nodes)",
+    );
+  }
+  if (nodes.size === 0) {
+    throw new MermaidParseError("no nodes parsed from the flowchart");
+  }
+
+  const graph: Graph = {
+    nodes: Array.from(nodes.values()),
+    direction,
+  };
+  if (groups.length > 0) graph.groups = groups;
+  if (edges.length > 0) graph.edges = edges;
+  return graph;
+}
+
+/** Index of the first arrow in a segment (for chain splitting). */
+function splitIndex(s: string): number {
+  let best = -1;
+  for (const arrow of ARROWS) {
+    const m = arrow.re.exec(s);
+    if (m && (best === -1 || m.index < best)) best = m.index;
+  }
+  return best === -1 ? s.length : best;
+}

packages/mcp/src/lib/drawio-presets.ts

@@ -0,0 +1,151 @@
+// Semantic color/line presets for the graph tools (issue #425, stage 3). The
+// PALETTE is DATA (packages/mcp/data/drawio-presets.json), not code: a node
+// `kind` maps to a { fillColor, strokeColor, fontColor } slot and an edge `kind`
+// maps to line-style props, per named preset (`default` / `dark` /
+// `colorblind-safe`). This module only loads that data and turns a slot into a
+// draw.io style fragment. The INVARIANT of the graph tools is that the model
+// never sees a style string — it names a `kind`, the server picks the slot.
+//
+// Loading mirrors drawio-shapes.ts: the JSON is read once via `import.meta.url`
+// relative to the built module. That is why this module (and drawio-graph.ts
+// which imports it) is reached ONLY through client.ts's ESM build and never
+// value-imported into the zod-agnostic tool-specs.ts (which the in-app server
+// type-checks under module:commonjs, where `import.meta` is a TS1343 error).
+
+import { readFileSync } from "node:fs";
+
+/** A node color slot: the three draw.io color values for a `kind`. */
+export interface NodeSlot {
+  fillColor: string;
+  strokeColor: string;
+  fontColor: string;
+}
+
+/** An edge line style: the extra style props appended for an edge `kind`. */
+export interface EdgeStyle {
+  props: string;
+}
+
+export interface PresetData {
+  canvasDark: boolean;
+  okabeIto?: string[];
+  nodes: Record<string, NodeSlot>;
+  edges: Record<string, EdgeStyle>;
+  edgeDefault: { strokeColor: string; fontColor: string };
+  group: { strokeColor: string; fontColor: string };
+}
+
+interface PresetsFile {
+  presets: Record<string, PresetData>;
+}
+
+/** The three shipped preset names. */
+export const PRESET_NAMES = ["default", "dark", "colorblind-safe"] as const;
+export type PresetName = (typeof PRESET_NAMES)[number];
+
+/** Every node `kind` the base palette defines (also the generic-shape kinds). */
+export const NODE_KINDS = [
+  "service",
+  "db",
+  "queue",
+  "gateway",
+  "error",
+  "external",
+  "security",
+] as const;
+export type NodeKind = (typeof NODE_KINDS)[number];
+
+/** Edge `kind`s the palette styles; anything else falls back to `sync`. */
+export const EDGE_KINDS = ["sync", "async", "error"] as const;
+export type EdgeKind = (typeof EDGE_KINDS)[number];
+
+let _presets: Record<string, PresetData> | null = null;
+
+function presetsPath(): URL {
+  // build/lib/drawio-presets.js -> ../../data/drawio-presets.json
+  return new URL("../../data/drawio-presets.json", import.meta.url);
+}
+
+/** Load + parse the bundled preset table once, then cache it. */
+export function loadPresets(): Record<string, PresetData> {
+  if (_presets) return _presets;
+  const json = readFileSync(presetsPath(), "utf-8");
+  const parsed = JSON.parse(json) as PresetsFile;
+  _presets = parsed.presets;
+  return _presets;
+}
+
+/** Resolve a preset by name, defaulting to `default` for an unknown name. */
+export function getPreset(name?: string): PresetData {
+  const presets = loadPresets();
+  if (name && presets[name]) return presets[name];
+  return presets["default"];
+}
+
+/** The slot for a node `kind` in a preset, falling back to `service`. */
+export function nodeSlot(preset: PresetData, kind?: string): NodeSlot {
+  if (kind && preset.nodes[kind]) return preset.nodes[kind];
+  return preset.nodes["service"];
+}
+
+/**
+ * Build the draw.io style string for a GENERIC (no-icon) node of a given kind.
+ * A rounded rectangle carrying the slot's fill/stroke/font. `whiteSpace=wrap`
+ * and `html=1` let a long label wrap inside the shape (the assembler also sizes
+ * the shape to the label, so the linter's label-overflow warning never fires).
+ */
+export function genericNodeStyle(preset: PresetData, kind?: string): string {
+  const s = nodeSlot(preset, kind);
+  return (
+    `rounded=1;whiteSpace=wrap;html=1;` +
+    `fillColor=${s.fillColor};strokeColor=${s.strokeColor};fontColor=${s.fontColor};`
+  );
+}
+
+/**
+ * Overlay the preset's node slot colors onto a resolved ICON style-string
+ * (from the shape catalog). An AWS/Azure icon carries its OWN mandatory
+ * fill/stroke (the category color / white outline) that MUST NOT be recolored,
+ * so for an icon we only ensure a readable fontColor when the preset is dark;
+ * otherwise the icon style is returned verbatim. Keeping the icon's own colors
+ * is deliberate: recoloring an AWS service icon breaks its category semantics.
+ */
+export function iconNodeStyle(preset: PresetData, iconStyle: string): string {
+  if (!preset.canvasDark) return iconStyle;
+  // On a dark canvas an icon's fontColor is usually a dark ink that vanishes;
+  // append a light fontColor (icons put their label BELOW the glyph, so this
+  // only affects the caption, never the glyph fill).
+  if (/fontColor=/.test(iconStyle)) {
+    return iconStyle.replace(/fontColor=[^;]*/, "fontColor=#e0e0e0");
+  }
+  return iconStyle + (iconStyle.endsWith(";") ? "" : ";") + "fontColor=#e0e0e0;";
+}
+
+/**
+ * Build the draw.io style for an edge of a given `kind`. Base is an orthogonal
+ * connector (edgeStyle=orthogonalEdgeStyle) with rounded corners and an open
+ * arrowhead, plus the preset's default stroke/font, then the kind's extra props
+ * (dashed / colored) overlaid. An unknown kind falls back to `sync` (solid).
+ */
+export function edgeStyle(preset: PresetData, kind?: string): string {
+  const k = kind && preset.edges[kind] ? kind : "sync";
+  const base =
+    `edgeStyle=orthogonalEdgeStyle;rounded=1;html=1;endArrow=open;` +
+    `strokeColor=${preset.edgeDefault.strokeColor};fontColor=${preset.edgeDefault.fontColor};`;
+  return base + preset.edges[k].props;
+}
+
+/**
+ * Group (container) style: ALWAYS transparent (`fillColor=none;container=1;`)
+ * per the spec, carrying the preset's group stroke/font. `dropTarget=1` marks it
+ * a drop target in the editor; `verticalAlign=top;align=left;spacingLeft=8;` puts
+ * the group label in the top-left like draw.io's own boundary containers.
+ */
+export function groupStyle(preset: PresetData): string {
+  return (
+    `rounded=0;whiteSpace=wrap;html=1;` +
+    `fillColor=none;container=1;dropTarget=1;collapsible=0;` +
+    `strokeColor=${preset.group.strokeColor};fontColor=${preset.group.fontColor};` +
+    `verticalAlign=top;align=left;spacingLeft=8;spacingTop=4;`
+  );
+}

packages/mcp/src/lib/filters.ts

@@ -83,16 +83,32 @@ export function filterComment(comment: any, markdownContent?: string) {
   };
 }
 
+// Map one server search hit to the MCP output contract (#443):
+//   { pageId, title, path, snippet, score }
+//
+// INVARIANT: the only page identifier exposed is `pageId` (the server `id`
+// UUID). The server also carries `slugId` — it is NEVER surfaced.
+//
+// GRACEFUL DEGRADATION: against a stock upstream server the opt-in lookup DTO
+// fields are stripped, so the response is the legacy FTS shape (no path/snippet/
+// score, a `highlight` + `rank` instead). We synthesize the contract from
+// whatever is present: `snippet` falls back to the FTS `highlight`, `score` to
+// the FTS `rank`, and `path` to [] (upstream has no path). This keeps the tool
+// usable even when the server has not been upgraded.
 export function filterSearchResult(result: any) {
   return {
-    id: result.id,
+    pageId: result.id,
     title: result.title,
-    parentPageId: result.parentPageId,
+    path: Array.isArray(result.path) ? result.path : [],
-    createdAt: result.createdAt,
+    snippet:
-    updatedAt: result.updatedAt,
+      typeof result.snippet === "string"
-    rank: result.rank,
+        ? result.snippet
-    highlight: result.highlight,
+        : (result.highlight ?? ""),
-    spaceId: result.space?.id,
+    score:
-    spaceName: result.space?.name,
+      typeof result.score === "number"
+        ? result.score
+        : typeof result.rank === "number"
+          ? result.rank
+          : 0,
   };
 }

packages/mcp/src/lib/page-lock.ts

@@ -10,6 +10,20 @@
 
 const chains = new Map<string, Promise<unknown>>();
 
+// Canonical UUID shape (versions 1–8, matching the `uuid` package's `validate`
+// that the server's isValidUUID uses). This is the SINGLE source of truth for
+// "is this a canonical page UUID?" in the MCP: client.ts's resolvePageId
+// imports isUuid from here to decide whether a pageId already IS a UUID (and so
+// needs no /pages/info round-trip). page.repo.ts treats any non-UUID pageId as
+// a slugId; a 10-char nanoid slugId never contains dashes, so it can never be
+// misread as a UUID here.
+export const UUID_RE =
+  /^[0-9a-f]{8}-[0-9a-f]{4}-[1-8][0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$/i;
+
+export function isUuid(value: string): boolean {
+  return typeof value === "string" && UUID_RE.test(value);
+}
+
 // The returned promise carries the real result/rejection of `fn` and MUST be
 // awaited/handled by the caller; only the internal chaining tail swallows
 // errors (purely to gate ordering).
@@ -17,6 +31,25 @@ export function withPageLock<T>(
   pageId: string,
   fn: () => Promise<T>,
 ): Promise<T> {
+  // STRUCTURAL INVARIANT (issue #449, "resolve-then-lock"): the mutex key MUST
+  // be the canonical page UUID, never a raw slugId. The whole write path relies
+  // on the lock key AND the CollabSession cache key being the resolved UUID
+  // (#260) — if a future write method forgot to call resolvePageId and locked
+  // under a slugId, two writes to the same page would take DIFFERENT mutex keys
+  // and silently lose serialization (clobbering live human edits). This was an
+  // invariant enforced only by comments/convention; assert it in CODE so the
+  // violation fails fast and loud at the lock instead of corrupting data in
+  // prod. The centralizing helper (mutatePageContent/replacePageContent) already
+  // guards a raw-input caller, but this backstop catches ANY path.
+  if (!isUuid(pageId)) {
+    throw new Error(
+      `withPageLock: key must be a canonical page UUID, got '${pageId}'. ` +
+        `The write path must resolvePageId(pageId) BEFORE locking so the ` +
+        `mutex/CollabSession cache key is the UUID (invariant "resolve-then-` +
+        `lock", #260/#449). A slugId or other non-UUID key would silently lose ` +
+        `per-page serialization.`,
+    );
+  }
   // Wait for the previous op on this page; swallow its error so a failure does
   // not poison the queue for the next caller.
   const prev = (chains.get(pageId) ?? Promise.resolve()).catch(() => {});

packages/mcp/src/lib/tree.ts

@@ -1,11 +1,30 @@
+/**
+ * Options for `buildPageTree`. Fully OPTIONAL so the existing call form
+ * `buildPageTree(nodes)` keeps its historic behaviour (lean `{id, slugId,
+ * title, children?}` output, no depth cut) unchanged.
+ *
+ *  - `shape: "getTree"` — emit the #443 `getTree` output node shape
+ *    `{pageId, title, children?, hasChildren?}` instead of the lean
+ *    `{id, slugId, title, children?}` shape. `slugId`/`icon`/`position` are
+ *    never exposed (INVARIANT: only the UUID `pageId` leaves the MCP layer).
+ *  - `maxDepth` — trim the built tree to this many levels (root nodes are
+ *    depth 1). Only meaningful together with `shape: "getTree"` (the lean shape
+ *    has no `hasChildren` to signal a cut). See the depth logic below.
+ */
+export interface BuildPageTreeOptions {
+  shape?: "lean" | "getTree";
+  maxDepth?: number;
+}
+
 /**
  * Pure tree-builder: turn a flat array of sidebar-style page nodes (as produced
  * by `enumerateSpacePages`) into a nested tree.
  *
  * Input: a flat array of nodes. Each node is expected to carry at least
- *   { id, slugId, title, position, parentPageId } (extra fields are ignored).
+ *   { id, slugId, title, position, parentPageId } (extra fields are ignored),
+ *   plus a server `hasChildren` boolean used by the `getTree` shape below.
  *
- * Output: an array of ROOT nodes, each shaped as
+ * Output (default / `shape: "lean"`): an array of ROOT nodes, each shaped as
  *   { id, slugId, title, children? }
  * where `children` is the array of child nodes (same shape, recursively). The
  * `children` key is OMITTED entirely when a node has no children — consistent
@@ -13,6 +32,14 @@
  * lean (nesting alone conveys the structure; parentPageId/position/hasChildren
  * are intentionally dropped from the output).
  *
+ * Output (`shape: "getTree"`, the #443 tool shape): each node is
+ *   { pageId, title, children?, hasChildren? }
+ * — the server `id` is exposed as `pageId` (never `slugId`/`icon`/`position`).
+ * `children` is omitted for leaves and for nodes trimmed by `maxDepth`.
+ * `hasChildren: true` is set ONLY on a node whose children exist on the server
+ * (per the flat item's `hasChildren`) but were CUT by `maxDepth`; on leaves and
+ * on fully-expanded interior nodes the field is omitted (see `maxDepth` below).
+ *
  * Linking rule: a node is attached as a child of `parentPageId` only when that
  * parent id is actually present in the input. Otherwise — including a null /
  * undefined `parentPageId`, or a parent that was capped out of the bounded walk
@@ -26,18 +53,42 @@
  * fractional-index ASCII keys (e.g. "a0", "a1"). Nodes with a missing/undefined
  * `position` sort last.
  *
+ * maxDepth (getTree shape only): the tree is built in FULL first, then trimmed
+ * on the way out. Root nodes are depth 1. `maxDepth: N` keeps nodes at depth
+ * <= N and drops the `children` of any node AT depth N. A node whose children
+ * were dropped this way gets `hasChildren: true` when it actually had children
+ * in the flat input (source of truth = the server `hasChildren` flag), so the
+ * caller knows it can descend further with a follow-up `rootPageId` call. An
+ * absent/undefined `maxDepth` means no cut (whole tree). `maxDepth <= 0` is
+ * treated as "no cut" (defensive; the tool schema clamps to >= 1).
+ *
  * Pure: no I/O, no network, deterministic.
  */
-export function buildPageTree(nodes: any[]): any[] {
+export function buildPageTree(
-  type OutputNode = {
+  nodes: any[],
+  options: BuildPageTreeOptions = {},
+): any[] {
+  const getTreeShape = options.shape === "getTree";
+  // A finite, positive cut only; anything else means "no cut".
+  const maxDepth =
+    typeof options.maxDepth === "number" &&
+    Number.isFinite(options.maxDepth) &&
+    options.maxDepth > 0
+      ? Math.floor(options.maxDepth)
+      : undefined;
+
+  type InternalNode = {
     id: string;
+    // Retained internally for shaping; never all emitted at once.
     slugId: any;
     title: any;
-    children?: OutputNode[];
+    hasServerChildren: boolean;
+    children?: InternalNode[];
   };
 
-  // Map id -> output node. Build the lean output shape up front.
+  // Map id -> internal node. Build up front; the output shape is projected at
-  const byId = new Map<string, OutputNode>();
+  // the very end so the maxDepth cut can consult `hasServerChildren`.
+  const byId = new Map<string, InternalNode>();
   // Preserve the original position string for sorting (kept off the output).
   const positionById = new Map<string, string | undefined>();
 
@@ -49,6 +100,7 @@ export function buildPageTree(nodes: any[]): any[] {
       id: node.id,
       slugId: node.slugId,
       title: node.title,
+      hasServerChildren: node.hasChildren === true,
     });
     positionById.set(node.id, node.position);
   }
@@ -90,5 +142,30 @@ export function buildPageTree(nodes: any[]): any[] {
   }
 
   roots.sort(byPosition);
-  return roots.map((id) => byId.get(id)!);
+  const rootNodes = roots.map((id) => byId.get(id)!);
+
+  // Project the internal nodes into the requested OUTPUT shape, applying the
+  // maxDepth cut for the getTree shape. `depth` is 1-based (roots = depth 1).
+  const project = (node: InternalNode, depth: number): any => {
+    if (getTreeShape) {
+      const out: any = { pageId: node.id, title: node.title };
+      const atCut = maxDepth !== undefined && depth >= maxDepth;
+      if (!atCut && node.children && node.children.length > 0) {
+        out.children = node.children.map((c) => project(c, depth + 1));
+      } else if (atCut && node.hasServerChildren) {
+        // Children exist on the server but were trimmed by maxDepth: signal it
+        // so the caller can descend with a follow-up rootPageId call.
+        out.hasChildren = true;
+      }
+      return out;
+    }
+    // Lean (historic) shape: cycle-safe, no depth cut, no hasChildren.
+    const out: any = { id: node.id, slugId: node.slugId, title: node.title };
+    if (node.children && node.children.length > 0) {
+      out.children = node.children.map((c) => project(c, depth + 1));
+    }
+    return out;
+  };
+
+  return rootNodes.map((n) => project(n, 1));
 }

packages/mcp/src/server-instructions.ts

@@ -40,8 +40,8 @@ import { SHARED_TOOL_SPECS, SharedToolSpec } from "./tool-specs.js";
  */
 export const ROUTING_PROSE =
   "Docmost editing guide — choose the tool by intent. The <tool_inventory> at the end lists every tool with a one-line purpose; the notes below are the routing hints for WHEN to reach for each.\n" +
-  "READ: find a page -> search (workspace-wide full-text); list -> listPages / listSpaces. Locate blocks and their ids CHEAPLY -> getOutline (compact top-level map; start here, not getPageJson). One block, for editing -> getNode (by attrs.id, or \"#<index>\" for tables, which carry no id) — returns MARKDOWN by default (comment anchors kept for safe write-back); pass format:\"json\" for the raw ProseMirror subtree. Find every occurrence of a string/regex ON a page (and where each is) -> searchInPage, NOT block-by-block getNode — it returns each hit's node ref + block index + context for a targeted comment. Whole page -> getPage (Markdown, lossy; inline <span data-comment-id> tags are comment anchors — markup, not text) or getPageJson (lossless ProseMirror with block ids). Hand a huge page (with images) to an external consumer without pulling it through the model context -> stashPage (returns a short-lived anonymous URL).\n" +
+  "READ: find a page by a fragment of a technical string (hostname/IP/ID like srv.local, 10.0.12, WB-MGE-30D86B) -> search — hybrid substring + full-text, returns each hit's location (path: root->parent titles) and a snippet around the match, so you rarely need a follow-up getPage; scope with spaceId or parentPageId (a subtree), titleOnly to match titles only. A space's page HIERARCHY (or one subtree) -> getTree (one request, complete, `{pageId,title,children?}`; rootPageId for a subtree, maxDepth to trim depth — a trimmed node gets hasChildren:true); prefer it over listPages tree:true (deprecated). Have a pageId, need WHERE-AM-I / what's around it (its breadcrumbs + direct children, metadata only) -> getPageContext (one call; parent = last breadcrumb, [] for a root page). list -> listPages / listSpaces. Locate blocks and their ids CHEAPLY -> getOutline (compact top-level map; start here, not getPageJson). One block, for editing -> getNode (by attrs.id, or \"#<index>\" for tables, which carry no id) — returns MARKDOWN by default (comment anchors kept for safe write-back); pass format:\"json\" for the raw ProseMirror subtree. Find every occurrence of a string/regex ON a page (and where each is) -> searchInPage, NOT block-by-block getNode — it returns each hit's node ref + block index + context for a targeted comment. Whole page -> getPage (Markdown, canonical for text; drops only block ids, resolved-comment anchors, and a fixed no-md-representation attr set: table spans/colwidth/bg, indent, callout.icon, orderedList.type, link internal/target/rel/class; inline <span data-comment-id> tags are comment anchors — markup, not text) or getPageJson (full ProseMirror with block ids, for those dropped attrs). Hand a huge page (with images) to an external consumer without pulling it through the model context -> stashPage (returns a short-lived anonymous URL).\n" +
-  "EDIT: fix wording/typos/numbers -> editPageText (find/replace inside blocks, no node id needed). Edit a block -> getNode(markdown) -> edit the markdown -> patchNode(markdown) (by attrs.id from getOutline; the markdown fragment may be several blocks — a 1->N section rewrite in one call, the first block keeps the id). Reach for patchNode's `node`-JSON only for fine attr/mark work; a table cell with spans/colors/fixed width -> the table tools (patchNode markdown refuses it). Add a block -> insertNode (markdown, before/after a block by attrs.id or by anchor text, or append; `node` for raw JSON or bare table structure). Remove a block -> deleteNode (by attrs.id). Tables -> tableGet / tableUpdateCell / tableInsertRow / tableDeleteRow (address by \"#<index>\" from getOutline; table nodes have no attrs.id). Images -> insertImage (add from a web URL) / replaceImage (swap an existing image). Draw.io diagrams -> drawioCreate (create from mxGraph XML and insert), drawioGet (read a diagram as mxGraph XML + a hash), drawioUpdate (replace a diagram; pass the hash from drawioGet as baseHash for optimistic locking); before authoring a diagram, drawioShapes (look up verified stencil style-strings so a shape name never renders as an empty box) and drawioGuide (on-demand authoring reference: skeleton/layout/containers/icons-aws/icons-azure), and pass layout:\"elk\" to drawioCreate/drawioUpdate to auto-place nodes. Footnotes -> insertFootnote. Bulk/structural rewrite -> updatePageJson (full ProseMirror replace) or updatePageMarkdown (full plain-Markdown body replace, re-imported — block ids regenerate); prefer the granular tools above to avoid resending the whole ~100KB+ document. Complex/scripted rewrite (multiple coordinated edits, renumbering) -> docmostTransform: write a JS `(doc, ctx) => doc` transform, preview the diff with dryRun (default), then apply with dryRun:false; ctx.helpers includes commentsToFootnotes for turning inline comments into numbered footnotes.\n" +
+  "EDIT: fix wording/typos/numbers -> editPageText (find/replace inside blocks, no node id needed). Edit a block -> getNode(markdown) -> edit the markdown -> patchNode(markdown) (by attrs.id from getOutline; the markdown fragment may be several blocks — a 1->N section rewrite in one call, the first block keeps the id). Reach for patchNode's `node`-JSON only for fine attr/mark work; a table cell with spans/colors/fixed width -> the table tools (patchNode markdown refuses it). Add a block -> insertNode (markdown, before/after a block by attrs.id or by anchor text, or append; `node` for raw JSON or bare table structure). Remove a block -> deleteNode (by attrs.id). Tables -> tableGet / tableUpdateCell / tableInsertRow / tableDeleteRow (address by \"#<index>\" from getOutline; table nodes have no attrs.id). Images -> insertImage (add from a web URL) / replaceImage (swap an existing image). Draw.io diagrams -> PREFER the high-level semantic tools that hide coordinates/styles: drawioFromGraph (architecture/cloud/network diagrams — describe nodes/groups/edges by kind+icon, the server picks layout, colors and verified icons; hints layer/sameLayerAs/pinned and layout:full|incremental|none) and drawioFromMermaid (standard flowcharts — write Mermaid, get an editable diagram). For targeted tweaks of an existing diagram use drawioEditCells (id-based add/update/delete with cascade delete + baseHash lock). Raw mxGraph XML via drawioCreate/drawioUpdate is the escape-hatch for exotic/wireframe diagrams; drawioGet reads a diagram as mxGraph XML + a hash (pass it as baseHash to drawioUpdate/drawioEditCells for optimistic locking). Before authoring raw XML, drawioShapes (look up verified stencil style-strings so a shape name never renders as an empty box) and drawioGuide (on-demand authoring reference: skeleton/layout/containers/icons-aws/icons-azure), and pass layout:\"elk\" to drawioCreate/drawioUpdate to auto-place nodes. Footnotes -> insertFootnote. Bulk/structural rewrite -> updatePageJson (full ProseMirror replace) or updatePageMarkdown (full plain-Markdown body replace, re-imported — block ids regenerate); prefer the granular tools above to avoid resending the whole ~100KB+ document. Complex/scripted rewrite (multiple coordinated edits, renumbering) -> docmostTransform: write a JS `(doc, ctx) => doc` transform, preview the diff with dryRun (default), then apply with dryRun:false; ctx.helpers includes commentsToFootnotes for turning inline comments into numbered footnotes.\n" +
   "PAGES: new -> createPage (Markdown). Rename (title only) -> renamePage. Move -> movePage. Delete -> deletePage (SOFT delete — the page goes to trash and is restorable; nothing is permanent). Copy/replace a page's whole content from another page (server-side, no document through the model) -> copyPageContent. Sharing -> sharePage / unsharePage / listShares; sharePage makes the page PUBLICLY accessible — do it only when explicitly asked.\n" +
   "COMMENTS: createComment is always inline and requires an EXACT selection — contiguous text from a single block, <=250 chars (fails rather than leaving an unanchored comment); reply to a thread via parentCommentId. Propose a concrete text fix for one-click human approval -> createComment with suggestedText (the exact plain-text replacement for the selection; the selection must then be UNIQUE in the page — extend it with context if needed); prefer this over editing directly when the change is subjective or needs the author's sign-off. Manage -> listComments, updateComment, resolveComment (resolve/reopen, reversible — prefer over delete to close), deleteComment, checkNewComments.\n" +
   "HISTORY: review what changed -> diffPageVersions (a historyId vs current, or two versions). List saved versions -> listPageHistory. Undo a bad edit -> restorePageVersion (writes a past version back as current; itself revertible). Export a page to self-contained Docmost Markdown (with comment anchors) -> exportPageMarkdown.";
@@ -82,6 +82,8 @@ const TOOL_FAMILY: Record<string, Family> = {
   // READ
   search: "READ",
   listPages: "READ",
+  getTree: "READ",
+  getPageContext: "READ",
   listSpaces: "READ",
   getOutline: "READ",
   getNode: "READ",
@@ -107,6 +109,9 @@ const TOOL_FAMILY: Record<string, Family> = {
   drawioGet: "EDIT",
   drawioCreate: "EDIT",
   drawioUpdate: "EDIT",
+  drawioEditCells: "EDIT",
+  drawioFromGraph: "EDIT",
+  drawioFromMermaid: "EDIT",
   drawioShapes: "EDIT",
   drawioGuide: "EDIT",
   docmostTransform: "EDIT",
@@ -152,7 +157,7 @@ export const INLINE_MCP_INVENTORY: ToolInventoryLine[] = [
   {
     name: "search",
     purpose:
-      "full-text search for pages and content across the whole workspace.",
+      "find pages by a fragment of a technical string (hybrid substring + full-text); returns each hit's path and a snippet.",
   },
   {
     name: "docmostTransform",

packages/mcp/src/tool-specs.ts

@@ -63,6 +63,8 @@ export type DocmostClientLike = Pick<
   | 'getSpaces'
   | 'listShares'
   | 'listPages'
+  | 'getTree'
+  | 'getPageContext'
   | 'getPage'
   | 'getPageJson'
   | 'getOutline'
@@ -98,6 +100,9 @@ export type DocmostClientLike = Pick<
   | 'drawioGet'
   | 'drawioCreate'
   | 'drawioUpdate'
+  | 'drawioEditCells'
+  | 'drawioFromGraph'
+  | 'drawioFromMermaid'
   | 'createComment'
   | 'resolveComment'
 >;
@@ -877,11 +882,17 @@ export const SHARED_TOOL_SPECS = {
     inAppKey: 'getPage',
     description:
       'Fetch a single page as Markdown by its id. Returns the page title and ' +
-      'its Markdown content. The Markdown conversion is LOSSY (block ids, exact ' +
+      'its Markdown content. The converter is canonical (round-trips text and ' +
-      'table/callout structure are approximated); for a lossless representation ' +
+      'block structure), so this is sufficient for text edits; use the ' +
-      'use the lossless page-JSON read tool. Inline <span data-comment-id> tags in the markdown ' +
+      'page-JSON read tool only when you need what Markdown cannot carry. The ' +
-      'are comment highlight anchors (also present for RESOLVED threads) — ' +
+      'Markdown drops exactly: (1) block ids (not visible in Markdown); ' +
-      'treat them as markup, not page text.',
+      '(2) resolved-comment anchors (hidden here; only active <span ' +
+      'data-comment-id> anchors remain); (3) a fixed set of attributes with no ' +
+      'Markdown representation — table-cell colspan/rowspan/colwidth/' +
+      'backgroundColor/backgroundColorName, heading/paragraph indent, ' +
+      'callout.icon, orderedList.type, and link internal/target/rel/class. ' +
+      'Inline <span data-comment-id> tags in the markdown are comment highlight ' +
+      'anchors — treat them as markup, not page text.',
     tier: 'core',
     catalogLine: 'getPage — fetch a page as Markdown by its id.',
     // Reconciled: MCP's stricter .min(1) kept; in-app's more-informative
@@ -911,11 +922,13 @@ export const SHARED_TOOL_SPECS = {
     description:
       'List the most recent pages (ordered by updatedAt, descending), ' +
       'optionally scoped to a single space. Returns a bounded list (default ' +
-      '50, max 100) — use search for lookups in large spaces. Pass tree:true ' +
+      '50, max 100) — use search for lookups in large spaces. tree:true (with ' +
-      "(with spaceId) to instead get the space's full page hierarchy as a " +
+      "spaceId) returns the space's full page hierarchy as a nested tree, but " +
-      'nested tree.',
+      'is DEPRECATED — use getTree instead (leaner nodes, plus rootPageId / ' +
+      'maxDepth).',
     tier: 'core',
-    catalogLine: "listPages — list recent pages, or a space's full page tree.",
+    catalogLine:
+      "listPages — list recent pages (tree:true is deprecated; use getTree for the hierarchy).",
     buildShape: (z) => ({
       spaceId: z
         .string()
@@ -948,6 +961,79 @@ export const SHARED_TOOL_SPECS = {
       ),
   },
 
+  getTree: {
+    mcpName: 'getTree',
+    inAppKey: 'getTree',
+    description:
+      "Get a space's page hierarchy (or one subtree) as a nested tree in a " +
+      'SINGLE request — completely and without loss. Each node is ' +
+      '`{ pageId, title, children? }`; children are ordered as in the sidebar. ' +
+      'Pass rootPageId to return only that page and its descendants (exactly ' +
+      'one root). Pass maxDepth to trim depth and save tokens (root nodes are ' +
+      'depth 1, so maxDepth:1 returns only the roots); a node whose children ' +
+      'were trimmed carries `hasChildren:true` so you can descend later with ' +
+      'getTree(rootPageId=that page). Prefer this over listPages tree:true.',
+    tier: 'core',
+    catalogLine:
+      "getTree — a space's page hierarchy (or a subtree) as a nested tree in one request.",
+    buildShape: (z) => ({
+      spaceId: z
+        .string()
+        .min(1)
+        .describe('The id of the space whose page tree to return.'),
+      rootPageId: z
+        .string()
+        .optional()
+        .describe(
+          'Optional page id: return only this page and its descendants (one root).',
+        ),
+      maxDepth: z
+        .number()
+        .int()
+        .min(1)
+        .optional()
+        .describe(
+          'Optional depth cap (roots are depth 1). maxDepth:1 returns only the ' +
+            'roots; trimmed nodes carry hasChildren:true.',
+        ),
+    }),
+    execute: (client, { spaceId, rootPageId, maxDepth }) =>
+      client.getTree(
+        spaceId as string,
+        rootPageId as string | undefined,
+        maxDepth as number | undefined,
+      ),
+  },
+
+  getPageContext: {
+    mcpName: 'getPageContext',
+    inAppKey: 'getPageContext',
+    description:
+      'Given a pageId, get its LOCATION and immediate surroundings (metadata ' +
+      'only, no page content) in one call — answers "where am I / what is ' +
+      "around this page\". Returns `{ page: { pageId, title, spaceId }, " +
+      'breadcrumbs: [{ pageId, title }], children: [{ pageId, title, ' +
+      'hasChildren }] }`. `breadcrumbs` is the ancestor chain from the space ' +
+      'root down to the PARENT (the parent is its last element; a root page ' +
+      'has `breadcrumbs: []`). `children` are the direct children in sidebar ' +
+      'order, each flagged `hasChildren` so you know which can be expanded ' +
+      '(descend with getTree(rootPageId=that child) or another getPageContext). ' +
+      'Ids, titles and child order are consistent with getTree.',
+    tier: 'core',
+    catalogLine:
+      'getPageContext — a page’s breadcrumbs + direct children (where-am-I) in one call.',
+    buildShape: (z) => ({
+      pageId: z
+        .string()
+        .min(1)
+        .describe(
+          'The id of the page to locate (a pageId/UUID, or a slugId from a URL).',
+        ),
+    }),
+    execute: (client, { pageId }) =>
+      client.getPageContext(pageId as string),
+  },
+
   createPage: {
     mcpName: 'createPage',
     inAppKey: 'createPage',
@@ -1240,13 +1326,17 @@ export const SHARED_TOOL_SPECS = {
     inAppKey: 'exportPageMarkdown',
     // CANONICAL: the MCP copy (a strict superset of the terse in-app wording).
     description:
-      'Export a page to a single self-contained, lossless Docmost-flavoured ' +
+      'Export a page to a single self-contained Docmost-flavoured Markdown ' +
-      'Markdown file (custom extensions): YAML-free meta header, body with ' +
+      'file (custom extensions): YAML-free meta header, body with inline ' +
-      'inline comment anchors and diagrams, and a trailing comments-thread ' +
+      'comment anchors (resolved ones kept) and diagrams, and a trailing ' +
-      'block. Designed for a download -> edit body -> page-Markdown import ' +
+      'comments-thread block. Designed for a download -> edit body -> ' +
-      'round-trip that preserves everything, including comment highlights. ' +
+      'page-Markdown import round-trip; block ids regenerate and comment ' +
-      'Comment THREADS are preserved in the file but are not re-pushed to the ' +
+      'THREADS, though kept in the file, are not re-pushed to the server on ' +
-      'server on import.',
+      'import. The round-trip SILENTLY DROPS a fixed set of attributes with no ' +
+      'Markdown representation — table-cell merge spans (colspan/rowspan), ' +
+      'colwidth, backgroundColor/backgroundColorName, heading/paragraph indent, ' +
+      'callout.icon, orderedList.type, and link internal/target/rel/class. Use ' +
+      'the page-JSON tools if those must survive.',
     tier: 'deferred',
     catalogLine:
       'exportPageMarkdown — export a page to self-contained Markdown (body + comments).',
@@ -1926,6 +2016,234 @@ export const SHARED_TOOL_SPECS = {
       ),
   },
 
+  drawioEditCells: {
+    mcpName: 'drawioEditCells',
+    inAppKey: 'drawioEditCells',
+    description:
+      'Make TARGETED, id-based edits to an existing draw.io diagram instead of ' +
+      'resending the whole XML (a full-XML diff is fragile — draw.io reorders ' +
+      'attributes). `operations` is an ordered list of: ' +
+      '{ op:"add", xml:"<mxCell .../>" } (append a new cell), ' +
+      '{ op:"update", cellId:"n3", xml:"<mxCell id=\\"n3\\" .../>" } (replace that ' +
+      'cell; the id MUST stay the same), or { op:"delete", cellId:"n5" } — a ' +
+      'delete CASCADES to the cell\'s container children AND to every edge whose ' +
+      'source/target is deleted. Ids are STABLE across edits so diffs stay ' +
+      'meaningful. `baseHash` is MANDATORY: pass the hash from the drawioGet you ' +
+      'based the edit on; if the diagram changed since, the edit is refused with ' +
+      'a conflict error — re-read with drawioGet and retry. The edited model goes ' +
+      'through the same lint + quality-warning pipeline as drawioUpdate. `node` is ' +
+      'the drawio node attrs.id or "#<index>". Use this to tweak a diagram (move ' +
+      'or restyle a few cells, add/remove nodes); to (re)generate a whole diagram ' +
+      'from a description use drawioFromGraph.' +
+      DRAWIO_HARD_RULES,
+    tier: 'deferred',
+    catalogLine:
+      'drawioEditCells — id-based add/update/delete edits to a draw.io diagram (cascade delete).',
+    buildShape: (z) => ({
+      pageId: z.string().min(1),
+      node: z
+        .string()
+        .min(1)
+        .describe('The drawio node attrs.id, or "#<index>" for a top-level block.'),
+      operations: z
+        .array(
+          z.object({
+            op: z.enum(['add', 'update', 'delete']),
+            cellId: z
+              .string()
+              .optional()
+              .describe('Target cell id (required for update/delete).'),
+            xml: z
+              .string()
+              .optional()
+              .describe('The <mxCell> element (required for add/update).'),
+          }),
+        )
+        .describe('Ordered add/update/delete operations keyed by cell id.'),
+      baseHash: z
+        .string()
+        .min(1)
+        .describe('The meta.hash from the drawioGet this edit is based on.'),
+    }),
+    execute: (client, { pageId, node, operations, baseHash }) =>
+      client.drawioEditCells(
+        pageId as string,
+        node as string,
+        operations as any,
+        baseHash as string,
+      ),
+  },
+
+  drawioFromGraph: {
+    mcpName: 'drawioFromGraph',
+    inAppKey: 'drawioFromGraph',
+    description:
+      'Build a draw.io diagram from a SEMANTIC graph — you describe nodes, groups ' +
+      'and edges by MEANING and the server picks every coordinate, color and icon ' +
+      'so the whole class of layout/icon mistakes (overlaps, edges through shapes, ' +
+      'empty-box stencils) cannot happen. This is the PREFERRED tool for ' +
+      'architecture / cloud / network diagrams. `graph` = { nodes:[{ id, label, ' +
+      'kind?, icon?, group?, layer?, sameLayerAs?, pinned? }], groups?:[{ id, ' +
+      'label, kind? }], edges?:[{ from, to, label?, kind? }] }. Node `kind` picks ' +
+      'a palette color (service/db/queue/gateway/error/external/security); `icon` ' +
+      '(e.g. "aws:lambda", "aws:dynamodb", "azure:cosmos") resolves to the exact ' +
+      'verified stencil — an unknown icon degrades to a labelled generic shape, ' +
+      'never an empty box. Edge `kind` sets the line style (sync=solid, ' +
+      'async=dashed, error=red-dashed). Groups are TRANSPARENT containers. ' +
+      '`direction` (LR/RL/TB/BT) and `preset` (default/dark/colorblind-safe) tune ' +
+      'the layout/palette. Layout hints: `layer` (column index), `sameLayerAs` ' +
+      '(align two nodes), `pinned:{x,y}` (fix a node). `layout`: "full" (default, ' +
+      'auto-place everything), "incremental" (with `node`: keep the existing ' +
+      'diagram\'s coordinates, place only new cells), "none" (no auto-layout). The ' +
+      'result reports { iconsResolved, iconsMissing } so you can verify all icons ' +
+      'resolved. For standard flowcharts you can also write Mermaid and call ' +
+      'drawioFromMermaid; for exotic/wireframe diagrams use raw XML via drawioCreate.',
+    tier: 'deferred',
+    catalogLine:
+      'drawioFromGraph — build a draw.io diagram from a semantic node/group/edge graph (server picks layout+icons).',
+    buildShape: (z) => {
+      const node = z.object({
+        id: z.string().min(1),
+        label: z.string().min(1),
+        kind: z
+          .string()
+          .optional()
+          .describe(
+            'Palette slot: service/db/queue/gateway/error/external/security.',
+          ),
+        icon: z
+          .string()
+          .optional()
+          .describe('Icon ref, e.g. "aws:lambda", "aws:dynamodb", "azure:cosmos".'),
+        group: z.string().optional().describe('Id of the group (container) it sits in.'),
+        layer: z.number().optional().describe('Layer/column index hint (>=0).'),
+        sameLayerAs: z
+          .string()
+          .optional()
+          .describe('Put this node in the same layer as another node id.'),
+        pinned: z
+          .object({ x: z.number(), y: z.number() })
+          .optional()
+          .describe('Fix the node at these exact coordinates.'),
+      });
+      const group = z.object({
+        id: z.string().min(1),
+        label: z.string().min(1),
+        kind: z.string().optional(),
+      });
+      const edge = z.object({
+        from: z.string().min(1),
+        to: z.string().min(1),
+        label: z.string().optional(),
+        kind: z
+          .string()
+          .optional()
+          .describe('sync (solid), async (dashed), error (red-dashed).'),
+      });
+      return {
+        pageId: z.string().min(1),
+        graph: z
+          .object({
+            nodes: z.array(node),
+            groups: z.array(group).optional(),
+            edges: z.array(edge).optional(),
+            direction: z.enum(['LR', 'RL', 'TB', 'BT']).optional(),
+            preset: z.enum(['default', 'dark', 'colorblind-safe']).optional(),
+          })
+          .describe('The semantic graph: nodes, groups, edges.'),
+        position: z
+          .enum(['before', 'after', 'append'])
+          .describe('Where to insert relative to the anchor.'),
+        anchorNodeId: z.string().optional().describe('Anchor block id (for before/after).'),
+        anchorText: z.string().optional().describe('Anchor text fragment (for before/after).'),
+        direction: z
+          .enum(['LR', 'RL', 'TB', 'BT'])
+          .optional()
+          .describe('Layout direction (overrides graph.direction).'),
+        preset: z
+          .enum(['default', 'dark', 'colorblind-safe'])
+          .optional()
+          .describe('Color preset (overrides graph.preset).'),
+        layout: z
+          .enum(['none', 'full', 'incremental'])
+          .optional()
+          .describe(
+            '"full" (default) auto-places all; "incremental" (with node) keeps ' +
+              'existing coords and places only new cells; "none" no auto-layout.',
+          ),
+        node: z
+          .string()
+          .optional()
+          .describe(
+            'An existing diagram to (re)build into — required for layout:"incremental".',
+          ),
+      };
+    },
+    execute: (
+      client,
+      { pageId, graph, position, anchorNodeId, anchorText, direction, preset, layout, node },
+    ) =>
+      client.drawioFromGraph(
+        pageId as string,
+        {
+          position: position as 'before' | 'after' | 'append',
+          anchorNodeId: anchorNodeId as string | undefined,
+          anchorText: anchorText as string | undefined,
+        },
+        graph as any,
+        direction as 'LR' | 'RL' | 'TB' | 'BT' | undefined,
+        preset as string | undefined,
+        layout as 'none' | 'full' | 'incremental' | undefined,
+        node as string | undefined,
+      ),
+  },
+
+  drawioFromMermaid: {
+    mcpName: 'drawioFromMermaid',
+    inAppKey: 'drawioFromMermaid',
+    description:
+      'Convert Mermaid `flowchart` text into an EDITABLE draw.io diagram (LLMs ' +
+      'write Mermaid reliably). Best for STANDARD flowcharts/decision trees: ' +
+      'write the mermaid, the server parses it (pure parser — no browser/CLI), ' +
+      'maps it to the same semantic pipeline as drawioFromGraph, and inserts a ' +
+      'real draw.io diagram you can then refine with drawioEditCells. Node shapes ' +
+      'map to palette colors (a `{decision}` -> yellow, a `[(db)]` -> green, etc.); ' +
+      '`subgraph … end` becomes a transparent group; dotted `-.->` edges become ' +
+      'dashed. ONLY flowchart/graph is supported — for sequence/class diagrams, or ' +
+      'for cloud/architecture diagrams with real service icons, use drawioFromGraph ' +
+      'instead. `where` positions the block like insertNode.',
+    tier: 'deferred',
+    catalogLine:
+      'drawioFromMermaid — turn Mermaid flowchart text into an editable draw.io diagram.',
+    buildShape: (z) => ({
+      pageId: z.string().min(1),
+      mermaid: z
+        .string()
+        .min(1)
+        .describe('Mermaid flowchart source (flowchart/graph LR|TB|...).'),
+      position: z
+        .enum(['before', 'after', 'append'])
+        .describe('Where to insert relative to the anchor.'),
+      anchorNodeId: z.string().optional().describe('Anchor block id (for before/after).'),
+      anchorText: z.string().optional().describe('Anchor text fragment (for before/after).'),
+      preset: z
+        .enum(['default', 'dark', 'colorblind-safe'])
+        .optional()
+        .describe('Color preset.'),
+    }),
+    execute: (client, { pageId, mermaid, position, anchorNodeId, anchorText, preset }) =>
+      client.drawioFromMermaid(
+        pageId as string,
+        {
+          position: position as 'before' | 'after' | 'append',
+          anchorNodeId: anchorNodeId as string | undefined,
+          anchorText: anchorText as string | undefined,
+        },
+        mermaid as string,
+        preset as string | undefined,
+      ),
+  },
+
   drawioShapes: {
     mcpName: 'drawioShapes',
     inAppKey: 'drawioShapes',

packages/mcp/test-e2e.mjs

@@ -316,7 +316,8 @@ async function main() {
         const [idA, idB, idC] = seedIds;
 
         // patchNode: replace the middle paragraph; siblings' ids must be unchanged.
-        await client.patchNode(nid, idB, mkPara(idB, "Bravo PATCHED."));
+        // #413 XOR input: the raw ProseMirror node goes under the `node` key.
+        await client.patchNode(nid, idB, { node: mkPara(idB, "Bravo PATCHED.") });
         await new Promise((r) => setTimeout(r, 16000));
         const afterPatch = (await client.getPageJson(nid)).content;
         const patchText = JSON.stringify(afterPatch);
@@ -327,7 +328,7 @@ async function main() {
         // insertNode: place a new block after the first paragraph.
         await client.insertNode(
           nid,
-          mkPara("nodeops-ins", "Inserted paragraph."),
+          { node: mkPara("nodeops-ins", "Inserted paragraph.") },
           { position: "after", anchorNodeId: idA },
         );
         await new Promise((r) => setTimeout(r, 16000));

packages/mcp/test/mock/drawio-graph-tools.test.mjs

@@ -0,0 +1,272 @@
+// Contract tests for the stage-3 drawio client methods (issue #425):
+// drawioEditCells / drawioFromGraph / drawioFromMermaid. Same seam-override
+// pattern as drawio-tools.test.mjs: a DocmostClient subclass stubs the I/O seams
+// so the tool logic runs without a live Docmost / collab socket.
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { DocmostClient } from "../../build/client.js";
+import {
+  buildDrawioSvg,
+  normalizeXml,
+  mxHash,
+  decodeDrawioSvg,
+  parseCells,
+} from "../../build/lib/drawio-xml.js";
+
+const DRAWIO_SCHEMA_ATTRS = new Set([
+  "src", "title", "alt", "width", "height", "size", "aspectRatio", "align", "attachmentId",
+]);
+function applyDrawioSchemaDrop(node) {
+  if (!node || typeof node !== "object") return;
+  if (node.type === "drawio" && node.attrs && typeof node.attrs === "object") {
+    for (const key of Object.keys(node.attrs))
+      if (!DRAWIO_SCHEMA_ATTRS.has(key)) delete node.attrs[key];
+  }
+  if (Array.isArray(node.content)) for (const c of node.content) applyDrawioSchemaDrop(c);
+}
+
+function svgFor(model, bbox = { width: 400, height: 300 }) {
+  return buildDrawioSvg(normalizeXml(model), "<g/>", bbox);
+}
+
+function makeClient({ pageDoc, attachmentSvg } = {}) {
+  const calls = { uploads: [], mutations: [] };
+  class TestClient extends DocmostClient {
+    async ensureAuthenticated() {}
+    async getCollabTokenWithReauth() {
+      return "collab-token";
+    }
+    async resolvePageId(pageId) {
+      return `uuid-${pageId}`;
+    }
+    async getPageRaw(pageId) {
+      return {
+        id: pageId, slugId: "s", title: "P", spaceId: "sp",
+        content: pageDoc ?? { type: "doc", content: [] },
+      };
+    }
+    async uploadAttachmentBuffer(pageId, buffer, fileName) {
+      const id = `att-${calls.uploads.length + 1}`;
+      calls.uploads.push({ pageId, fileName, svg: buffer.toString("utf-8") });
+      return { id, fileName, fileSize: buffer.length };
+    }
+    async fetchAttachmentText() {
+      return attachmentSvg;
+    }
+    mutatePage(pageId, token, apiUrl, transform) {
+      const clone = structuredClone(pageDoc ?? { type: "doc", content: [] });
+      const doc = transform(clone);
+      if (doc) applyDrawioSchemaDrop(doc);
+      calls.mutations.push({ pageId, doc });
+      return Promise.resolve({ doc, verify: { changed: doc != null } });
+    }
+  }
+  const client = new TestClient("http://127.0.0.1:1/api", "e@x.com", "pw");
+  return { client, calls };
+}
+
+function findDrawio(node, acc = []) {
+  if (!node || typeof node !== "object") return acc;
+  if (node.type === "drawio") acc.push(node);
+  if (Array.isArray(node.content)) for (const c of node.content) findDrawio(c, acc);
+  return acc;
+}
+
+// A stored diagram: a group with two children and an edge.
+const STORED =
+  "<mxGraphModel><root><mxCell id=\"0\"/><mxCell id=\"1\" parent=\"0\"/>" +
+  '<mxCell id="grp" value="G" style="container=1;fillColor=none;" vertex="1" parent="1">' +
+  '<mxGeometry x="0" y="0" width="300" height="200" as="geometry"/></mxCell>' +
+  '<mxCell id="a" value="A" style="rounded=1;" vertex="1" parent="grp">' +
+  '<mxGeometry x="10" y="10" width="80" height="40" as="geometry"/></mxCell>' +
+  '<mxCell id="b" value="B" style="rounded=1;" vertex="1" parent="grp">' +
+  '<mxGeometry x="10" y="90" width="80" height="40" as="geometry"/></mxCell>' +
+  '<mxCell id="e" style="" edge="1" parent="grp" source="a" target="b">' +
+  '<mxGeometry relative="1" as="geometry"/></mxCell>' +
+  "</root></mxGraphModel>";
+
+function drawioPageDoc() {
+  return {
+    type: "doc",
+    content: [
+      {
+        type: "drawio",
+        attrs: {
+          id: "d1", src: "/api/files/att-1/diagram.drawio.svg",
+          attachmentId: "att-1", width: 400, height: 300,
+        },
+      },
+    ],
+  };
+}
+
+// --- drawioEditCells --------------------------------------------------------
+
+test("drawioEditCells: applies ops and repoints the node (current baseHash)", async () => {
+  const { client, calls } = makeClient({
+    pageDoc: drawioPageDoc(),
+    attachmentSvg: svgFor(STORED),
+  });
+  const baseHash = mxHash(normalizeXml(STORED));
+  const res = await client.drawioEditCells(
+    "page1",
+    "d1",
+    [
+      {
+        op: "update",
+        cellId: "a",
+        xml:
+          '<mxCell id="a" value="Renamed" style="rounded=1;" vertex="1" parent="grp">' +
+          '<mxGeometry x="10" y="10" width="80" height="40" as="geometry"/></mxCell>',
+      },
+    ],
+    baseHash,
+  );
+  assert.equal(res.success, true);
+  assert.equal(calls.uploads.length, 1);
+  const written = decodeDrawioSvg(calls.uploads[0].svg);
+  const cells = parseCells(written);
+  assert.equal(cells.find((c) => c.id === "a").value, "Renamed");
+  assert.equal(cells.find((c) => c.id === "b").value, "B"); // untouched
+  const n = findDrawio(calls.mutations[0].doc)[0];
+  // The stub numbers uploads from 1; this edit is the first upload -> att-1.
+  assert.equal(n.attrs.attachmentId, "att-1");
+});
+
+test("drawioEditCells: delete of the container cascades to children + edge", async () => {
+  const { client, calls } = makeClient({
+    pageDoc: drawioPageDoc(),
+    attachmentSvg: svgFor(STORED),
+  });
+  const baseHash = mxHash(normalizeXml(STORED));
+  await client.drawioEditCells("page1", "d1", [{ op: "delete", cellId: "grp" }], baseHash);
+  const written = decodeDrawioSvg(calls.uploads[0].svg);
+  const ids = parseCells(written).filter((c) => c.id !== "0" && c.id !== "1").map((c) => c.id);
+  assert.deepEqual(ids, [], "grp + a + b + edge all cascaded away");
+});
+
+test("drawioEditCells: stale baseHash -> conflict, no upload", async () => {
+  const { client, calls } = makeClient({
+    pageDoc: drawioPageDoc(),
+    attachmentSvg: svgFor(STORED),
+  });
+  await assert.rejects(
+    () => client.drawioEditCells("page1", "d1", [{ op: "delete", cellId: "a" }], "stale"),
+    /conflict/,
+  );
+  assert.equal(calls.uploads.length, 0);
+});
+
+test("drawioEditCells: baseHash is mandatory", async () => {
+  const { client } = makeClient({ pageDoc: drawioPageDoc(), attachmentSvg: svgFor(STORED) });
+  await assert.rejects(
+    () => client.drawioEditCells("page1", "d1", [{ op: "delete", cellId: "a" }], ""),
+    /baseHash is mandatory/,
+  );
+});
+
+// --- drawioFromGraph --------------------------------------------------------
+
+test("drawioFromGraph: builds a diagram from a graph and inserts a node", async () => {
+  const pageDoc = { type: "doc", content: [{ type: "paragraph", attrs: { id: "p1" }, content: [] }] };
+  const { client, calls } = makeClient({ pageDoc });
+  const res = await client.drawioFromGraph(
+    "page1",
+    { position: "append" },
+    {
+      nodes: [
+        { id: "api", label: "API", kind: "gateway", icon: "aws:api_gateway", group: "vpc" },
+        { id: "fn", label: "Handler", kind: "service", icon: "aws:lambda", group: "vpc" },
+        { id: "db", label: "Orders", kind: "db", icon: "aws:dynamodb" },
+      ],
+      groups: [{ id: "vpc", label: "VPC" }],
+      edges: [{ from: "api", to: "fn", kind: "sync" }, { from: "fn", to: "db", kind: "async" }],
+    },
+    "LR",
+    "default",
+  );
+  assert.equal(res.success, true);
+  assert.equal(res.nodeId, "#1");
+  assert.equal(res.iconsMissing.length, 0, `unresolved: ${res.iconsMissing}`);
+  assert.equal(res.iconsResolved, 3);
+  // The uploaded model decodes back and carries the group + nodes.
+  const written = decodeDrawioSvg(calls.uploads[0].svg);
+  const cells = parseCells(written);
+  assert.ok(cells.some((c) => c.id === "vpc"));
+  assert.ok(cells.some((c) => c.id === "api"));
+  // Group is transparent.
+  const vpc = cells.find((c) => c.id === "vpc");
+  assert.equal(vpc.styleMap.fillColor, "none");
+  assert.equal(vpc.styleMap.container, "1");
+});
+
+test("drawioFromGraph: an invalid graph throws before any upload", async () => {
+  const { client, calls } = makeClient({ pageDoc: { type: "doc", content: [] } });
+  await assert.rejects(
+    () => client.drawioFromGraph("page1", { position: "append" }, { nodes: [] }),
+    /non-empty/,
+  );
+  assert.equal(calls.uploads.length, 0);
+});
+
+test("drawioFromGraph incremental into an existing node keeps prior coords", async () => {
+  // The stored diagram has a,b at known coords; add a new node c incrementally.
+  const { client, calls } = makeClient({
+    pageDoc: drawioPageDoc(),
+    attachmentSvg: svgFor(STORED),
+  });
+  const res = await client.drawioFromGraph(
+    "page1",
+    { position: "append" },
+    {
+      nodes: [
+        { id: "a", label: "A" },
+        { id: "b", label: "B" },
+        { id: "c", label: "C new" },
+      ],
+      edges: [{ from: "b", to: "c" }],
+    },
+    undefined,
+    undefined,
+    "incremental",
+    "d1", // target the existing diagram
+  );
+  assert.equal(res.success, true);
+  const written = decodeDrawioSvg(calls.uploads[0].svg);
+  const cells = parseCells(written);
+  const a = cells.find((c) => c.id === "a");
+  const b = cells.find((c) => c.id === "b");
+  // Existing coords preserved (the stored a/b absolute coords from STORED).
+  assert.equal(a.geometry.x, 10);
+  assert.equal(a.geometry.y, 10);
+  assert.equal(b.geometry.x, 10);
+  assert.equal(b.geometry.y, 90);
+  assert.ok(cells.some((c) => c.id === "c"), "new node c added");
+});
+
+// --- drawioFromMermaid ------------------------------------------------------
+
+test("drawioFromMermaid: converts a flowchart and inserts a diagram", async () => {
+  const pageDoc = { type: "doc", content: [{ type: "paragraph", attrs: { id: "p1" }, content: [] }] };
+  const { client, calls } = makeClient({ pageDoc });
+  const res = await client.drawioFromMermaid(
+    "page1",
+    { position: "append" },
+    "flowchart LR\n A[Start] --> B{Choose}\n B -->|yes| C[Done]\n B -->|no| D[Stop]",
+  );
+  assert.equal(res.success, true);
+  const written = decodeDrawioSvg(calls.uploads[0].svg);
+  const cells = parseCells(written);
+  for (const id of ["A", "B", "C", "D"]) {
+    assert.ok(cells.some((c) => c.id === id), `node ${id} present`);
+  }
+});
+
+test("drawioFromMermaid: a non-flowchart is rejected, no upload", async () => {
+  const { client, calls } = makeClient({ pageDoc: { type: "doc", content: [] } });
+  await assert.rejects(
+    () => client.drawioFromMermaid("page1", { position: "append" }, "sequenceDiagram\n A->>B: x"),
+    /only 'flowchart'\/'graph' is supported/,
+  );
+  assert.equal(calls.uploads.length, 0);
+});

packages/mcp/test/mock/get-page-context.test.mjs

@@ -0,0 +1,375 @@
+// Mock-HTTP tests for DocmostClient.getPageContext — the #443 "where am I /
+// what's around" read tool. A local http.createServer stands in for Docmost
+// (same harness style as pagination-cursor.test.mjs) so everything is
+// deterministic and offline.
+//
+// Contract pinned here:
+//   - Two requests: POST /pages/breadcrumbs (ancestor chain root->page, page
+//     INCLUDED as the LAST element) + listSidebarPages (direct children).
+//   - Split: last chain element -> `page`; the rest (root->parent) ->
+//     `breadcrumbs`. A ROOT page (chain length 1) -> breadcrumbs: [].
+//   - children: {pageId, title, hasChildren} in sidebar order.
+//   - INVARIANT: only the UUID `pageId` is exposed, never `slugId`.
+//   - A slugId input is resolved via /pages/info first (adds one request); a
+//     UUID input short-circuits (stays at two requests).
+//   - A bad/inaccessible pageId throws a CLEAR error, not an empty object.
+import { test, after } from "node:test";
+import assert from "node:assert/strict";
+import http from "node:http";
+import { DocmostClient } from "../../build/client.js";
+
+function readBody(req) {
+  return new Promise((resolve) => {
+    let raw = "";
+    req.on("data", (chunk) => {
+      raw += chunk;
+    });
+    req.on("end", () => resolve(raw));
+  });
+}
+
+function startServer(handler) {
+  return new Promise((resolve) => {
+    const server = http.createServer(handler);
+    server.listen(0, "127.0.0.1", () => {
+      const { port } = server.address();
+      resolve({ server, baseURL: `http://127.0.0.1:${port}/api` });
+    });
+  });
+}
+
+function closeServer(server) {
+  return new Promise((resolve) => server.close(resolve));
+}
+
+function sendJson(res, status, obj, extraHeaders = {}) {
+  res.writeHead(status, { "Content-Type": "application/json", ...extraHeaders });
+  res.end(JSON.stringify(obj));
+}
+
+const openServers = [];
+async function spawn(handler) {
+  const { server, baseURL } = await startServer(handler);
+  openServers.push(server);
+  return { server, baseURL };
+}
+
+after(async () => {
+  await Promise.all(openServers.map((s) => closeServer(s)));
+});
+
+function handleLogin(req, res) {
+  if (req.url === "/api/auth/login") {
+    sendJson(res, 200, { success: true }, {
+      "Set-Cookie": "authToken=t; Path=/; HttpOnly",
+    });
+    return true;
+  }
+  return false;
+}
+
+// Two real UUIDs so resolvePageId short-circuits (no /pages/info round-trip).
+const ROOT_UUID = "00000000-0000-4000-8000-000000000001";
+const MID_UUID = "00000000-0000-4000-8000-000000000002";
+const PAGE_UUID = "00000000-0000-4000-8000-000000000003";
+const CHILD_A = "00000000-0000-4000-8000-00000000000a";
+const CHILD_B = "00000000-0000-4000-8000-00000000000b";
+
+// Build a breadcrumbs response as the server sends it: root->page order, page
+// LAST, wrapped in the {data,success} envelope. slugId/icon/position are present
+// on the wire (they must NOT leak into the tool output).
+function breadcrumbsEnvelope(chain) {
+  return { success: true, data: chain };
+}
+
+// -----------------------------------------------------------------------------
+// 1) 3rd-level page: page = last chain element; breadcrumbs = the two ancestors
+//    root->parent; children mapped {pageId,title,hasChildren} in order; no leak;
+//    exactly two requests for a UUID input.
+// -----------------------------------------------------------------------------
+test("getPageContext: 3rd-level page splits chain, maps children, no slugId leak, 2 requests", async () => {
+  let breadcrumbReqs = 0;
+  let sidebarReqs = 0;
+  let infoReqs = 0;
+  let breadcrumbBody = null;
+
+  const { baseURL } = await spawn(async (req, res) => {
+    const raw = await readBody(req);
+    if (handleLogin(req, res)) return;
+    if (req.url === "/api/pages/info") {
+      infoReqs++;
+      sendJson(res, 404, {});
+      return;
+    }
+    if (req.url === "/api/pages/breadcrumbs") {
+      breadcrumbReqs++;
+      breadcrumbBody = JSON.parse(raw || "{}");
+      // root -> parent -> page (page LAST). slugId/icon/position on the wire.
+      sendJson(
+        res,
+        200,
+        breadcrumbsEnvelope([
+          { id: ROOT_UUID, slugId: "rootSlug", title: "Infrastructure", spaceId: "sp1", position: "a", icon: null, parentPageId: null, hasChildren: true },
+          { id: MID_UUID, slugId: "midSlug", title: "Datacenter A", spaceId: "sp1", position: "a", icon: null, parentPageId: ROOT_UUID, hasChildren: true },
+          { id: PAGE_UUID, slugId: "pageSlug", title: "Rack 12", spaceId: "sp1", position: "b", icon: null, parentPageId: MID_UUID, hasChildren: true },
+        ]),
+      );
+      return;
+    }
+    if (req.url === "/api/pages/sidebar-pages") {
+      sidebarReqs++;
+      const body = JSON.parse(raw || "{}");
+      assert.equal(body.pageId, PAGE_UUID, "children scoped to the page UUID");
+      assert.equal(body.spaceId, "sp1", "children scoped to the page's space");
+      sendJson(res, 200, {
+        success: true,
+        data: {
+          items: [
+            { id: CHILD_A, slugId: "aSlug", title: "Servers", parentPageId: PAGE_UUID, hasChildren: true, position: "a" },
+            { id: CHILD_B, slugId: "bSlug", title: "Network", parentPageId: PAGE_UUID, hasChildren: false, position: "b" },
+          ],
+          meta: { hasNextPage: false, nextCursor: null },
+        },
+      });
+      return;
+    }
+    sendJson(res, 404, {});
+  });
+
+  const client = new DocmostClient(baseURL, "user@example.com", "pw");
+  const result = await client.getPageContext(PAGE_UUID);
+
+  assert.equal(infoReqs, 0, "UUID input short-circuits resolvePageId (no /pages/info)");
+  assert.equal(breadcrumbReqs, 1, "exactly one breadcrumbs request");
+  assert.equal(sidebarReqs, 1, "exactly one sidebar request");
+  assert.deepEqual(breadcrumbBody, { pageId: PAGE_UUID }, "breadcrumbs posts the UUID");
+
+  // page = the LAST chain element.
+  assert.deepEqual(result.page, {
+    pageId: PAGE_UUID,
+    title: "Rack 12",
+    spaceId: "sp1",
+  });
+  // breadcrumbs = root->parent (the chain minus the page itself).
+  assert.deepEqual(result.breadcrumbs, [
+    { pageId: ROOT_UUID, title: "Infrastructure" },
+    { pageId: MID_UUID, title: "Datacenter A" },
+  ]);
+  // children mapped in order, hasChildren coerced to boolean.
+  assert.deepEqual(result.children, [
+    { pageId: CHILD_A, title: "Servers", hasChildren: true },
+    { pageId: CHILD_B, title: "Network", hasChildren: false },
+  ]);
+
+  // No slugId anywhere in the output.
+  const dump = JSON.stringify(result);
+  assert.ok(!dump.includes("Slug"), "no slugId leaks into the output");
+  assert.ok(!/\bslugId\b/.test(dump), "no slugId key in the output");
+});
+
+// -----------------------------------------------------------------------------
+// 2) ROOT page: chain has ONE element (the page itself) -> breadcrumbs: [].
+// -----------------------------------------------------------------------------
+test("getPageContext: a root page has breadcrumbs: []", async () => {
+  const { baseURL } = await spawn(async (req, res) => {
+    const raw = await readBody(req);
+    if (handleLogin(req, res)) return;
+    if (req.url === "/api/pages/breadcrumbs") {
+      // A root page: the CTE returns only the page itself.
+      sendJson(
+        res,
+        200,
+        breadcrumbsEnvelope([
+          { id: ROOT_UUID, slugId: "rootSlug", title: "Infrastructure", spaceId: "sp1", parentPageId: null, hasChildren: false },
+        ]),
+      );
+      return;
+    }
+    if (req.url === "/api/pages/sidebar-pages") {
+      sendJson(res, 200, {
+        success: true,
+        data: { items: [], meta: { hasNextPage: false, nextCursor: null } },
+      });
+      return;
+    }
+    sendJson(res, 404, {});
+  });
+
+  const client = new DocmostClient(baseURL, "user@example.com", "pw");
+  const result = await client.getPageContext(ROOT_UUID);
+
+  assert.deepEqual(result.page, {
+    pageId: ROOT_UUID,
+    title: "Infrastructure",
+    spaceId: "sp1",
+  });
+  assert.deepEqual(result.breadcrumbs, [], "root page: no ancestors");
+  assert.deepEqual(result.children, [], "no children");
+});
+
+// -----------------------------------------------------------------------------
+// 3) A slugId input is resolved via /pages/info first (one extra request), then
+//    breadcrumbs/sidebar use the resolved UUID.
+// -----------------------------------------------------------------------------
+test("getPageContext: a slugId input is resolved via /pages/info", async () => {
+  let infoReqs = 0;
+  let infoBody = null;
+  let breadcrumbBody = null;
+
+  const { baseURL } = await spawn(async (req, res) => {
+    const raw = await readBody(req);
+    if (handleLogin(req, res)) return;
+    if (req.url === "/api/pages/info") {
+      infoReqs++;
+      infoBody = JSON.parse(raw || "{}");
+      // getPageRaw: slugId -> canonical UUID.
+      sendJson(res, 200, {
+        success: true,
+        data: { id: PAGE_UUID, slugId: "pageSlug", title: "Rack 12", spaceId: "sp1" },
+      });
+      return;
+    }
+    if (req.url === "/api/pages/breadcrumbs") {
+      breadcrumbBody = JSON.parse(raw || "{}");
+      sendJson(
+        res,
+        200,
+        breadcrumbsEnvelope([
+          { id: ROOT_UUID, slugId: "rootSlug", title: "Infrastructure", spaceId: "sp1", parentPageId: null },
+          { id: PAGE_UUID, slugId: "pageSlug", title: "Rack 12", spaceId: "sp1", parentPageId: ROOT_UUID },
+        ]),
+      );
+      return;
+    }
+    if (req.url === "/api/pages/sidebar-pages") {
+      sendJson(res, 200, {
+        success: true,
+        data: { items: [], meta: { hasNextPage: false, nextCursor: null } },
+      });
+      return;
+    }
+    sendJson(res, 404, {});
+  });
+
+  const client = new DocmostClient(baseURL, "user@example.com", "pw");
+  const result = await client.getPageContext("pageSlug");
+
+  assert.equal(infoReqs, 1, "slugId resolved via one /pages/info");
+  assert.deepEqual(infoBody, { pageId: "pageSlug" }, "resolve posts the raw slugId");
+  assert.deepEqual(
+    breadcrumbBody,
+    { pageId: PAGE_UUID },
+    "breadcrumbs posts the RESOLVED uuid, not the slugId",
+  );
+  assert.equal(result.page.pageId, PAGE_UUID, "page.pageId is the UUID");
+  assert.deepEqual(result.breadcrumbs, [
+    { pageId: ROOT_UUID, title: "Infrastructure" },
+  ]);
+});
+
+// -----------------------------------------------------------------------------
+// 4) >20 children: cursor pagination returns ALL of them, no dupes (regression
+//    on the #442 bug class — getPageContext must not re-introduce a cap).
+// -----------------------------------------------------------------------------
+test("getPageContext: a page with >20 children returns ALL of them (no cap, no dupes)", async () => {
+  // 45 children spread over three cursor pages.
+  const all = Array.from({ length: 45 }, (_, i) => ({
+    id: `child-${i}`,
+    slugId: `slug-${i}`,
+    title: `Child ${i}`,
+    parentPageId: PAGE_UUID,
+    hasChildren: i % 2 === 0,
+  }));
+  const PAGES = {
+    "": { items: all.slice(0, 20), nextCursor: "c1" },
+    c1: { items: all.slice(20, 40), nextCursor: "c2" },
+    c2: { items: all.slice(40), nextCursor: null },
+  };
+
+  const { baseURL } = await spawn(async (req, res) => {
+    const raw = await readBody(req);
+    if (handleLogin(req, res)) return;
+    if (req.url === "/api/pages/breadcrumbs") {
+      sendJson(
+        res,
+        200,
+        breadcrumbsEnvelope([
+          { id: PAGE_UUID, slugId: "pageSlug", title: "Big Parent", spaceId: "sp1", parentPageId: null },
+        ]),
+      );
+      return;
+    }
+    if (req.url === "/api/pages/sidebar-pages") {
+      const body = JSON.parse(raw || "{}");
+      const page = PAGES[body.cursor ?? ""] ?? { items: [], nextCursor: null };
+      sendJson(res, 200, {
+        success: true,
+        data: {
+          items: page.items,
+          meta: { hasNextPage: page.nextCursor != null, nextCursor: page.nextCursor },
+        },
+      });
+      return;
+    }
+    sendJson(res, 404, {});
+  });
+
+  const client = new DocmostClient(baseURL, "user@example.com", "pw");
+  const result = await client.getPageContext(PAGE_UUID);
+
+  assert.equal(result.children.length, 45, "all 45 children returned");
+  const ids = result.children.map((c) => c.pageId);
+  assert.equal(new Set(ids).size, 45, "no duplicate children");
+  assert.deepEqual(ids, all.map((c) => c.id), "children in server order across cursor pages");
+  assert.equal(result.children[0].hasChildren, true, "hasChildren preserved (child 0)");
+  assert.equal(result.children[1].hasChildren, false, "hasChildren preserved (child 1)");
+});
+
+// -----------------------------------------------------------------------------
+// 5) A nonexistent / inaccessible pageId -> a CLEAR error, NOT an empty object.
+// -----------------------------------------------------------------------------
+test("getPageContext: a bad/inaccessible pageId throws a clear error (not {})", async () => {
+  const { baseURL } = await spawn(async (req, res) => {
+    await readBody(req);
+    if (handleLogin(req, res)) return;
+    if (req.url === "/api/pages/breadcrumbs") {
+      // Server rejects an unknown/forbidden page.
+      sendJson(res, 404, { message: "Page not found" });
+      return;
+    }
+    sendJson(res, 404, {});
+  });
+
+  const client = new DocmostClient(baseURL, "user@example.com", "pw");
+  await assert.rejects(
+    () => client.getPageContext(PAGE_UUID),
+    (err) => {
+      assert.ok(err instanceof Error, "throws an Error");
+      return true;
+    },
+    "a 404 from breadcrumbs propagates as a thrown error, not a hollow {}",
+  );
+});
+
+// -----------------------------------------------------------------------------
+// 6) An empty breadcrumbs chain (should never happen — the endpoint always
+//    includes the page itself) is treated as not-found, not a hollow {page:...}.
+// -----------------------------------------------------------------------------
+test("getPageContext: an empty breadcrumbs chain throws (defensive)", async () => {
+  const { baseURL } = await spawn(async (req, res) => {
+    await readBody(req);
+    if (handleLogin(req, res)) return;
+    if (req.url === "/api/pages/breadcrumbs") {
+      sendJson(res, 200, breadcrumbsEnvelope([]));
+      return;
+    }
+    sendJson(res, 404, {});
+  });
+
+  const client = new DocmostClient(baseURL, "user@example.com", "pw");
+  await assert.rejects(
+    () => client.getPageContext(PAGE_UUID),
+    /not found or inaccessible/,
+    "an empty chain is a clear error, not {}",
+  );
+});

packages/mcp/test/mock/getpage-conversion-cache.test.mjs

@@ -0,0 +1,251 @@
+// Mock-HTTP integration tests for the getPage conversion cache (issue #479).
+// A local http.createServer stands in for Docmost (same harness style as
+// get-page-context.test.mjs) so everything is deterministic and offline.
+//
+// Verifies end-to-end through the real client that:
+//   - the FIRST getPage of a page is a MISS (mcp_getpage_cache_misses_total)
+//     and converts the content (the server's convert-representative counter);
+//   - a SECOND getPage of the same (pageId, updatedAt) is a HIT
+//     (mcp_getpage_cache_hits_total) and returns BYTE-IDENTICAL output while
+//     skipping the conversion;
+//   - a changed updatedAt is a fresh key -> MISS again;
+//   - the returned shape still resolves page + subpages.
+import { test, after, mock } from "node:test";
+import assert from "node:assert/strict";
+import http from "node:http";
+import { DocmostClient } from "../../build/client.js";
+
+function readBody(req) {
+  return new Promise((resolve) => {
+    let raw = "";
+    req.on("data", (chunk) => (raw += chunk));
+    req.on("end", () => resolve(raw));
+  });
+}
+
+function sendJson(res, status, obj, extraHeaders = {}) {
+  res.writeHead(status, { "Content-Type": "application/json", ...extraHeaders });
+  res.end(JSON.stringify(obj));
+}
+
+const openServers = [];
+after(async () => {
+  await Promise.all(openServers.map((s) => new Promise((r) => s.close(r))));
+});
+
+const PAGE_UUID = "00000000-0000-4000-8000-000000000010";
+const SPACE_UUID = "00000000-0000-4000-8000-0000000000aa";
+const CHILD_UUID = "00000000-0000-4000-8000-0000000000bb";
+
+// A small ProseMirror doc so the converter produces non-trivial markdown.
+function makeDoc(text) {
+  return {
+    type: "doc",
+    content: [
+      {
+        type: "paragraph",
+        content: [{ type: "text", text }],
+      },
+    ],
+  };
+}
+
+// state.info counts /pages/info hits; state.updatedAt / state.text drive the
+// content+version returned; state.sidebar counts sidebar-pages hits;
+// state.subpages (when set) drives the child list the sidebar endpoint returns,
+// so a test can vary the live subpages across two reads of the same page.
+function spawn(state) {
+  return new Promise((resolve) => {
+    const server = http.createServer(async (req, res) => {
+      await readBody(req);
+      if (req.url === "/api/auth/login") {
+        return sendJson(res, 200, { success: true }, {
+          "Set-Cookie": "authToken=t; Path=/; HttpOnly",
+        });
+      }
+      if (req.url === "/api/pages/info") {
+        state.info++;
+        return sendJson(res, 200, {
+          success: true,
+          data: {
+            id: PAGE_UUID,
+            slugId: "slug123456",
+            title: "Cached Page",
+            parentPageId: null,
+            spaceId: SPACE_UUID,
+            updatedAt: state.updatedAt,
+            content: makeDoc(state.text),
+          },
+        });
+      }
+      if (req.url === "/api/pages/sidebar-pages") {
+        state.sidebar++;
+        const items = state.subpages ?? [
+          { id: CHILD_UUID, title: "Child", hasChildren: false },
+        ];
+        return sendJson(res, 200, {
+          success: true,
+          data: {
+            items,
+            meta: { hasNextPage: false, nextCursor: null },
+          },
+        });
+      }
+      return sendJson(res, 404, { message: "not found" });
+    });
+    server.listen(0, "127.0.0.1", () => {
+      openServers.push(server);
+      resolve(`http://127.0.0.1:${server.address().port}/api`);
+    });
+  });
+}
+
+function makeClient(baseURL, metrics) {
+  return new DocmostClient({
+    apiUrl: baseURL,
+    getToken: async () => "access",
+    onMetric: (name, value) => {
+      metrics[name] = (metrics[name] ?? 0) + value;
+    },
+  });
+}
+
+test("first read MISS, second read HIT with byte-identical output; convert runs once", async () => {
+  const state = { info: 0, sidebar: 0, updatedAt: "2026-01-01T00:00:00Z", text: "Hello world" };
+  const baseURL = await spawn(state);
+  const metrics = {};
+  const client = makeClient(baseURL, metrics);
+
+  const first = await client.getPage(PAGE_UUID);
+  assert.equal(metrics["mcp_getpage_cache_misses_total"], 1, "first read is a miss");
+  assert.equal(metrics["mcp_getpage_cache_hits_total"] ?? 0, 0, "no hit yet");
+
+  const second = await client.getPage(PAGE_UUID);
+  assert.equal(metrics["mcp_getpage_cache_hits_total"], 1, "second read is a hit");
+  assert.equal(metrics["mcp_getpage_cache_misses_total"], 1, "still one miss");
+
+  // BYTE-IDENTICAL: the cache only skips recomputation, never changes output.
+  assert.deepEqual(second, first, "cached result is identical to the uncached one");
+  assert.equal(
+    JSON.stringify(second),
+    JSON.stringify(first),
+    "serialized output is byte-identical",
+  );
+
+  // The page fetch + subpages fetch still happen every call (only the CPU
+  // conversion is cached); both reads hit /pages/info and sidebar-pages.
+  assert.equal(state.info, 2, "both reads still fetch /pages/info");
+  assert.equal(state.sidebar, 2, "both reads still fetch subpages");
+
+  // Shape sanity: content present, subpages resolved.
+  assert.equal(typeof second.data.content, "string");
+  assert.ok(second.data.content.includes("Hello world"));
+  assert.deepEqual(second.data.subpages, [{ id: CHILD_UUID, title: "Child" }]);
+});
+
+test("a changed updatedAt is a fresh key -> MISS again, with the NEW content", async () => {
+  const state = { info: 0, sidebar: 0, updatedAt: "2026-01-01T00:00:00Z", text: "Version one" };
+  const baseURL = await spawn(state);
+  const metrics = {};
+  const client = makeClient(baseURL, metrics);
+
+  const a = await client.getPage(PAGE_UUID); // miss
+  const b = await client.getPage(PAGE_UUID); // hit
+  assert.equal(metrics["mcp_getpage_cache_misses_total"], 1);
+  assert.equal(metrics["mcp_getpage_cache_hits_total"], 1);
+  assert.ok(a.data.content.includes("Version one"));
+
+  // The page changes: new updatedAt AND new content.
+  state.updatedAt = "2026-02-02T00:00:00Z";
+  state.text = "Version two";
+
+  const c = await client.getPage(PAGE_UUID); // miss on the new key
+  assert.equal(metrics["mcp_getpage_cache_misses_total"], 2, "changed version -> miss");
+  assert.equal(metrics["mcp_getpage_cache_hits_total"], 1, "no stale hit");
+  assert.ok(c.data.content.includes("Version two"), "the NEW content is served");
+  assert.ok(!c.data.content.includes("Version one"), "no stale markdown");
+
+  const d = await client.getPage(PAGE_UUID); // hit on the new key
+  assert.equal(metrics["mcp_getpage_cache_hits_total"], 2, "the new snapshot caches too");
+});
+
+test("a slugId read and a UUID read of the same page share one cache entry", async () => {
+  // resolvePageId maps the slugId -> UUID via /pages/info; the cache keys on the
+  // canonical UUID (resultData.id), so both inputs land on the same entry.
+  const state = { info: 0, sidebar: 0, updatedAt: "2026-01-01T00:00:00Z", text: "Shared" };
+  const baseURL = await spawn(state);
+  const metrics = {};
+  const client = makeClient(baseURL, metrics);
+
+  await client.getPage(PAGE_UUID);          // miss (keyed on UUID)
+  await client.getPage("slug123456");        // the server returns the same id -> HIT
+  assert.equal(metrics["mcp_getpage_cache_misses_total"], 1, "one conversion total");
+  assert.equal(metrics["mcp_getpage_cache_hits_total"], 1, "slugId read hits the UUID entry");
+});
+
+test("on a conversion HIT, the {{SUBPAGES}} block reflects the LIVE subpages, not the cached ones", async () => {
+  // The whole byte-identity guarantee: the cache stores the conversion output
+  // BEFORE the {{SUBPAGES}} substitution, so a re-read of an UNCHANGED page still
+  // splices the FRESH subpage list. The page body itself contains {{SUBPAGES}}
+  // (converts to a literal placeholder); getPage replaces it with the live list.
+  const CHILD_A = "00000000-0000-4000-8000-0000000000a1";
+  const CHILD_B = "00000000-0000-4000-8000-0000000000b2";
+  const state = {
+    info: 0,
+    sidebar: 0,
+    updatedAt: "2026-01-01T00:00:00Z", // FIXED across both reads -> conversion cache HIT
+    text: "Body before {{SUBPAGES}} body after",
+    subpages: [{ id: CHILD_A, title: "Alpha", hasChildren: false }],
+  };
+  const baseURL = await spawn(state);
+  const metrics = {};
+  const client = makeClient(baseURL, metrics);
+
+  // Read 1: MISS (converts). The substitution runs with list A.
+  const first = await client.getPage(PAGE_UUID);
+  assert.equal(metrics["mcp_getpage_cache_misses_total"], 1, "first read converts (miss)");
+  assert.ok(first.data.content.includes("[Alpha](page:" + CHILD_A + ")"), "list A spliced in");
+  assert.ok(!first.data.content.includes("{{SUBPAGES}}"), "placeholder consumed");
+
+  // The subpages change while the PAGE CONTENT/updatedAt do NOT: same conversion
+  // cache key -> a HIT that skips the CPU walk, but the live substitution must
+  // still run on the NEW list B.
+  state.subpages = [{ id: CHILD_B, title: "Beta", hasChildren: false }];
+
+  const second = await client.getPage(PAGE_UUID);
+  assert.equal(metrics["mcp_getpage_cache_hits_total"], 1, "second read is a conversion HIT");
+  assert.equal(metrics["mcp_getpage_cache_misses_total"], 1, "no second conversion");
+
+  // The cache did NOT freeze the subpages block: list B is present, list A gone.
+  assert.ok(second.data.content.includes("[Beta](page:" + CHILD_B + ")"), "live list B spliced in on a HIT");
+  assert.ok(!second.data.content.includes("Alpha"), "stale list A is NOT frozen into the output");
+  assert.deepEqual(second.data.subpages, [{ id: CHILD_B, title: "Beta" }], "subpages field reflects list B");
+});
+
+test("a cache HIT SKIPS the convertProseMirrorToMarkdown CPU walk (called once across MISS+HIT)", async () => {
+  // The single reason the cache exists: on a hit the expensive PM-tree walk must
+  // NOT run. The miss counter alone can't prove this — a broken hit branch that
+  // re-converted (same output, misses=1) would leave every other assert green.
+  // So spy directly on the conversion seam and assert the call COUNT.
+  const state = { info: 0, sidebar: 0, updatedAt: "2026-01-01T00:00:00Z", text: "Body text" };
+  const baseURL = await spawn(state);
+  const metrics = {};
+  const client = makeClient(baseURL, metrics);
+
+  // Spy on the seam that wraps convertProseMirrorToMarkdown; it still delegates,
+  // so output stays real and byte-identical — we only count invocations.
+  const spy = mock.method(client, "convertPageMarkdown");
+
+  await client.getPage(PAGE_UUID); // MISS -> converts once
+  assert.equal(spy.mock.callCount(), 1, "the miss converts exactly once");
+
+  await client.getPage(PAGE_UUID); // HIT -> must NOT convert again
+  assert.equal(
+    spy.mock.callCount(),
+    1,
+    "the hit skips the conversion: still exactly one call across MISS+HIT",
+  );
+  assert.equal(metrics["mcp_getpage_cache_hits_total"], 1, "and it was recorded as a hit");
+
+  spy.mock.restore();
+});

packages/mcp/test/mock/resolve-page-id-collab-doc-name.test.mjs

@@ -372,7 +372,11 @@ test("replaceImage opens by the resolved UUID AND keys its page lock by that UUI
   // single flush. This proves the flush actually executes queued callbacks, so
   // probeRan === false above means "blocked", not "the flush never ran anyone".
   let freeRan = false;
-  const freeDone = withPageLock(`page.free-${UUID}`, async () => {
+  // A DIFFERENT canonical UUID (unrelated to the page under test). withPageLock
+  // now asserts its key is a canonical UUID (#449), so the "free" probe key must
+  // also be a valid — but distinct — UUID, not a synthetic label.
+  const FREE_UUID = "99999999-9999-4999-8999-999999999999";
+  const freeDone = withPageLock(FREE_UUID, async () => {
     freeRan = true;
   });
   await new Promise((r) => setImmediate(r));

packages/mcp/test/unit/collab-session.test.mjs

@@ -323,7 +323,9 @@ test("MCP_COLLAB_SESSION_IDLE_MS=0 disables the cache (legacy provider-per-op)",
 });
 
 test("replaceImage-shaped flow: acquire under an EXTERNAL page lock does not deadlock and reuses one session", async () => {
-  const pageId = "page-lock";
+  // withPageLock now asserts a canonical UUID key (#449); this flow takes the
+  // real page lock (mirroring replaceImage), so the key must be a valid UUID.
+  const pageId = "77777777-7777-4777-8777-777777777777";
   // Mirror replaceImage: hold ONE withPageLock across scan (read-only) + write,
   // each going through the non-locking acquireCollabSession.
   const result = await withPageLock(pageId, async () => {

packages/mcp/test/unit/diff-guard-skips-recreate.test.mjs

@@ -0,0 +1,86 @@
+// Issue #464 — prove the size guard SKIPS the recreateTransform pipeline over
+// the cap, not merely that it returns "coarse". node:test's mock.module needs an
+// experimental flag the suite does not pass, so instead of a module spy we use a
+// deterministic BEHAVIORAL proxy that isolates the one variable — the guard:
+//
+//   Same over-cap pair, run twice:
+//     (a) default caps            -> guard trips  -> recreateTransform skipped,
+//     (b) caps raised above the doc -> guard OFF -> recreateTransform DOES run.
+//
+// The only code path that differs between (a) and (b) is whether
+// recreateTransform executes. recreateTransform on this pair is O(n²) and takes
+// SECONDS; the guarded path is a linear coarse diff taking milliseconds. So a
+// large (a)≪(b) time ratio can ONLY be explained by (a) skipping the transform.
+// This asserts the skip without depending on mock.module.
+import { test } from "node:test";
+import assert from "node:assert/strict";
+
+import { diffDocs } from "../../build/lib/diff.js";
+
+const t = (text) => ({ type: "text", text });
+const para = (text) => ({ type: "paragraph", content: text ? [t(text)] : [] });
+const doc = (children) => ({ type: "doc", content: children });
+function buildDoc(n, seed) {
+  return doc(
+    Array.from({ length: n }, (_, i) =>
+      para(Array.from({ length: 8 }, (_, w) => `${seed}${i}_${w}`).join(" ")),
+    ),
+  );
+}
+function clearEnv() {
+  delete process.env.MCP_DIFF_MAX_NODES;
+  delete process.env.MCP_DIFF_MAX_BYTES;
+}
+function timed(fn) {
+  const s = performance.now();
+  const out = fn();
+  return { out, ms: performance.now() - s };
+}
+
+// A 300-para (~600-node) pair: comfortably over the 150-node default, yet small
+// enough that the un-guarded recreateTransform still FINISHES (~1-3s) so the
+// test can time the contrast without hanging.
+const OLD = buildDoc(300, "a");
+const NEW = buildDoc(300, "b");
+
+test("guard skips recreateTransform over-cap (guarded run is far faster than un-guarded)", () => {
+  // (a) Guarded: default caps -> should short-circuit to coarse, near-instant.
+  clearEnv();
+  const guarded = timed(() => diffDocs(OLD, NEW));
+  assert.match(
+    guarded.out.markdown,
+    /coarse block-level diff/,
+    "guarded run must be coarse (guard tripped)",
+  );
+
+  // (b) Un-guarded: raise both caps above the doc so the precise path runs.
+  process.env.MCP_DIFF_MAX_NODES = "1000000";
+  process.env.MCP_DIFF_MAX_BYTES = "100000000";
+  let unguarded;
+  try {
+    unguarded = timed(() => diffDocs(OLD, NEW));
+  } finally {
+    clearEnv();
+  }
+  assert.doesNotMatch(
+    unguarded.out.markdown,
+    /coarse block-level diff/,
+    "with caps raised, the precise recreateTransform path runs",
+  );
+
+  // The precise run executed recreateTransform (O(n²)); the guarded run did not.
+  // Require a large speedup so the ONLY explanation is the skipped transform.
+  assert.ok(
+    guarded.ms * 5 < unguarded.ms,
+    `guarded (${guarded.ms.toFixed(1)}ms) must be >=5x faster than un-guarded ` +
+      `(${unguarded.ms.toFixed(1)}ms); a small gap would mean the transform still ran`,
+  );
+});
+
+test("guarded over-cap call stays within the ~200ms event-loop budget", () => {
+  clearEnv();
+  // Best-of-3 to shed GC/JIT noise; the guarded coarse path is a linear walk.
+  let best = Infinity;
+  for (let i = 0; i < 3; i++) best = Math.min(best, timed(() => diffDocs(OLD, NEW)).ms);
+  assert.ok(best < 200, `guarded over-cap diff must be <200ms, was ${best.toFixed(1)}ms`);
+});

packages/mcp/test/unit/diff-size-guard.test.mjs

@@ -0,0 +1,229 @@
+// Issue #464 — prod CPU-DoS pre-flight size guard for diffDocs.
+//
+// diffDocs synchronously calls recreateTransform (rfc6902) which is O(n·m) in
+// node count and O(w²) in per-run word count; on a large/heavily-changed doc it
+// pins the event loop for seconds-to-hours WITHOUT throwing. A pre-flight size
+// guard routes any doc over MCP_DIFF_MAX_NODES / MCP_DIFF_MAX_BYTES straight to
+// the coarse fallback (`fellBack:true`), so the sync block stays ~<200ms.
+//
+// These tests assert the BEHAVIOR of the guard (fast + coarse-mode + asymmetry +
+// env knobs). A sibling test (diff-guard-skips-recreate.test.mjs) proves
+// recreateTransform is skipped over the cap via a behavioral proxy (guarded run
+// is orders of magnitude faster than the same pair with the caps raised).
+import { test } from "node:test";
+import assert from "node:assert/strict";
+
+import { diffDocs } from "../../build/lib/diff.js";
+
+// ---------------------------------------------------------------------------
+// Builders
+// ---------------------------------------------------------------------------
+const t = (text) => ({ type: "text", text });
+const para = (text) => ({ type: "paragraph", content: text ? [t(text)] : [] });
+const doc = (children) => ({ type: "doc", content: children });
+
+/** A doc of `n` paragraphs whose words are seeded from `seed` (fully changeable). */
+function buildDoc(n, wordsPerPara, seed) {
+  const blocks = [];
+  for (let i = 0; i < n; i++) {
+    const words = [];
+    for (let w = 0; w < wordsPerPara; w++) words.push(`${seed}${i}_${w}`);
+    blocks.push(para(words.join(" ")));
+  }
+  return doc(blocks);
+}
+
+/** Reset the env knobs to their unset default between tests. */
+function clearEnv() {
+  delete process.env.MCP_DIFF_MAX_NODES;
+  delete process.env.MCP_DIFF_MAX_BYTES;
+}
+
+// ---------------------------------------------------------------------------
+// Over-threshold (by node count) -> FAST + coarse mode.
+// A fully re-written 600-para doc is the worst case that drove the incident;
+// with the guard it must return in well under the ~200ms budget and in coarse
+// mode. Without the guard this single call takes multiple SECONDS.
+// ---------------------------------------------------------------------------
+test("over-threshold doc falls back to coarse mode and returns fast", () => {
+  clearEnv();
+  // 600 paragraphs -> ~1200 nodes, far over the 150-node default.
+  const oldDoc = buildDoc(600, 8, "a");
+  const newDoc = buildDoc(600, 8, "b");
+
+  const start = performance.now();
+  const r = diffDocs(oldDoc, newDoc);
+  const elapsed = performance.now() - start;
+
+  // Coarse mode is signalled in the markdown note (fellBack path).
+  assert.match(
+    r.markdown,
+    /coarse block-level diff/,
+    "over-threshold pair must use the coarse fallback",
+  );
+  // Budget: the guard makes this near-instant. Generous 1s ceiling to avoid CI
+  // flake while still being ~10x under the multi-second un-guarded cost.
+  assert.ok(
+    elapsed < 1000,
+    `expected fast coarse fallback, took ${elapsed.toFixed(0)}ms`,
+  );
+  // Coarse diff still detects the wholesale change.
+  assert.ok(r.summary.inserted > 0 || r.summary.deleted > 0, "reports changes");
+});
+
+// ---------------------------------------------------------------------------
+// Under-threshold (small) doc -> precise diff, NOT coarse mode. No regression.
+// ---------------------------------------------------------------------------
+test("under-threshold doc uses the precise diff (no fallback note)", () => {
+  clearEnv();
+  const oldDoc = doc([para("Hello world")]);
+  const newDoc = doc([para("Hello brave world")]);
+  const r = diffDocs(oldDoc, newDoc);
+
+  assert.doesNotMatch(
+    r.markdown,
+    /coarse block-level diff/,
+    "a small doc must take the precise path",
+  );
+  // Precise word diff finds exactly the inserted word.
+  const ins = r.changes.find((c) => c.op === "insert");
+  assert.ok(ins && /brave/.test(ins.text), "precise diff isolates the inserted word");
+});
+
+// ---------------------------------------------------------------------------
+// Asymmetry: a small NEW doc vs a huge OLD doc (and vice versa) still explodes
+// rfc6902, so max(old,new) must trip the guard in BOTH directions.
+// ---------------------------------------------------------------------------
+test("asymmetric pair (huge old, tiny new) falls back to coarse", () => {
+  clearEnv();
+  const hugeOld = buildDoc(600, 8, "a");
+  const tinyNew = doc([para("just one line")]);
+  const r = diffDocs(hugeOld, tinyNew);
+  assert.match(r.markdown, /coarse block-level diff/, "huge-old side must trip the guard");
+});
+
+test("asymmetric pair (tiny old, huge new) falls back to coarse", () => {
+  clearEnv();
+  const tinyOld = doc([para("just one line")]);
+  const hugeNew = buildDoc(600, 8, "b");
+  const r = diffDocs(tinyOld, hugeNew);
+  assert.match(r.markdown, /coarse block-level diff/, "huge-new side must trip the guard");
+});
+
+// ---------------------------------------------------------------------------
+// Byte axis: a FEW nodes but a very large serialized size (long text runs) is
+// dangerous too (per-run word diff is O(words²)), so the byte cap must trip
+// independently of the node count.
+// ---------------------------------------------------------------------------
+test("node-light but byte-heavy doc falls back on the byte cap", () => {
+  clearEnv();
+  // 5 paragraphs (~11 nodes, well under the node cap) but each a very long run,
+  // pushing the serialized size far over the 12 KiB byte default.
+  const bigRun = (seed) =>
+    doc(
+      Array.from({ length: 5 }, (_, i) =>
+        para(Array.from({ length: 800 }, (_, w) => `${seed}${i}_${w}`).join(" ")),
+      ),
+    );
+  const oldDoc = bigRun("a");
+  const newDoc = bigRun("b");
+  // Sanity: node count is under the default node cap, so ONLY the byte cap can
+  // be what trips the guard here.
+  const nodeCount = (d) => {
+    let n = 0;
+    const v = (x) => {
+      if (!x || typeof x !== "object") return;
+      n++;
+      if (Array.isArray(x.content)) for (const c of x.content) v(c);
+    };
+    v(d);
+    return n;
+  };
+  assert.ok(nodeCount(oldDoc) < 150, "node count is under the node cap");
+  assert.ok(JSON.stringify(oldDoc).length > 12 * 1024, "serialized size is over the byte cap");
+
+  const r = diffDocs(oldDoc, newDoc);
+  assert.match(r.markdown, /coarse block-level diff/, "byte cap must trip independently");
+});
+
+// ---------------------------------------------------------------------------
+// Env override: a very low MCP_DIFF_MAX_NODES forces fallback on a tiny doc,
+// proving the knob is read fresh and actually gates the diff.
+// ---------------------------------------------------------------------------
+test("MCP_DIFF_MAX_NODES override forces fallback on a small doc", () => {
+  clearEnv();
+  const oldDoc = doc([para("Hello world")]);
+  const newDoc = doc([para("Hello brave world")]);
+
+  // Baseline: default caps -> precise diff.
+  assert.doesNotMatch(diffDocs(oldDoc, newDoc).markdown, /coarse block-level diff/);
+
+  // Knob set absurdly low -> even this 4-node doc trips the guard.
+  process.env.MCP_DIFF_MAX_NODES = "1";
+  try {
+    const r = diffDocs(oldDoc, newDoc);
+    assert.match(r.markdown, /coarse block-level diff/, "low node cap forces fallback");
+  } finally {
+    clearEnv();
+  }
+});
+
+test("MCP_DIFF_MAX_BYTES override forces fallback on a small doc", () => {
+  clearEnv();
+  const oldDoc = doc([para("Hello world")]);
+  const newDoc = doc([para("Hello brave world")]);
+
+  process.env.MCP_DIFF_MAX_BYTES = "1";
+  try {
+    const r = diffDocs(oldDoc, newDoc);
+    assert.match(r.markdown, /coarse block-level diff/, "low byte cap forces fallback");
+  } finally {
+    clearEnv();
+  }
+});
+
+// ---------------------------------------------------------------------------
+// Garbage / unset env values fall back to the DEFAULT (the guard can never be
+// accidentally disabled by a malformed knob). A small doc must still diff
+// precisely under a garbage cap.
+// ---------------------------------------------------------------------------
+test("garbage env values fall back to the default cap (guard not disabled)", () => {
+  clearEnv();
+  const oldDoc = doc([para("Hello world")]);
+  const newDoc = doc([para("Hello brave world")]);
+
+  for (const bad of ["not-a-number", "0", "-5", "", "NaN", "1e999"]) {
+    process.env.MCP_DIFF_MAX_NODES = bad;
+    process.env.MCP_DIFF_MAX_BYTES = bad;
+    // Under the DEFAULT caps this small doc is precise (garbage did not raise
+    // OR disable the cap). "1e999" -> parseInt yields 1 (finite) which is a
+    // valid low cap and would fall back; exclude that from the precise check.
+    const r = diffDocs(oldDoc, newDoc);
+    if (bad === "1e999") {
+      // parseInt("1e999",10) === 1 -> a legit low cap -> fallback. Guard active.
+      assert.match(r.markdown, /coarse block-level diff/);
+    } else {
+      assert.doesNotMatch(
+        r.markdown,
+        /coarse block-level diff/,
+        `garbage value ${JSON.stringify(bad)} must fall back to the default cap`,
+      );
+    }
+  }
+  clearEnv();
+});
+
+// ---------------------------------------------------------------------------
+// A large doc that trips the guard must still return the correct INTEGRITY
+// counts (computeIntegrity runs before the diff and is unaffected by fallback).
+// ---------------------------------------------------------------------------
+test("integrity counts are still correct on a guard-tripped (coarse) doc", () => {
+  clearEnv();
+  const image = { type: "image", attrs: { src: "/api/files/a.png" } };
+  const oldDoc = doc([image, ...buildDoc(600, 8, "a").content]);
+  const newDoc = doc([...buildDoc(600, 8, "b").content]); // image removed
+
+  const r = diffDocs(oldDoc, newDoc);
+  assert.match(r.markdown, /coarse block-level diff/, "large pair fell back");
+  assert.deepEqual(r.integrity.images, [1, 0], "integrity is computed regardless of fallback");
+});