Enquire MCP

v3.4.0 — Computes structural communities over a complete resource-bounded vault wikilink graph via greedy modularity optimization (single-phase Louvain). Returns exact total/eligible/returned community and member counts, membership-path JSON UTF-8 byte receipts, explicit truncation receipts, `modularity` (∈ [-0.5, 1]), `iterations`, `converged`, and bounded `communities[]` with id/size/sorted-members/representative. Membership is encoded once in `communities[].members`; no duplicate path→id map is emitted. Pure structural — no embeddings or LLM calls. The result is NOT cached — call once per s…

Return a note's immediate graph neighborhood in one call: outbound wikilinks (resolved), inbound backlinks (with count), and tag-cluster siblings (notes sharing ≥1 tag, excluding outbound/inbound). Replaces the read_note → backlinks → outbound → resolve_wikilink chain with a single round-trip — designed for RAG-style 'give the LLM enough context to reason about THIS note'.

List every link this note points to — wikilinks and (optionally) embeds, with each one's resolution status. Symmetric counterpart to obsidian_get_backlinks.

List notes ordered by most recent modification. Useful for picking up where work was left off.

Find every [[wikilink]] (and ![[embed]]) in the vault whose target does not resolve to a file. Useful as a vault-hygiene utility — broken links, typos, notes you intended to create.

v3.1.0 — HyDE retrieval (Gao et al 2023). Caller agent generates a synthetic answer to its own question, passes it as `hypothetical_answer`; the server embeds the answer (not the question) and retrieves against the answer-shaped vector. Typically beats raw-query embedding by +2-5 NDCG@10 on under-specified queries (e.g. "what did I learn about X" — the question vector is generic; the answer vector is topically anchored). Uses the same `.embed.db` as `obsidian_embeddings_search`. The agent SHOULD generate the hypothetical answer with no vault access (otherwise the loop is circular); 1-3 sentenc…

Comprehensive vault-hygiene check inspired by Karpathy's LLM-Wiki gist (gist.github.com/karpathy/442a6bf555914893e9891c11519de94f). Returns five buckets of findings in one call: orphans (no inbound + no outbound), broken wikilinks, stub pages (under N words), stale pages (frontmatter `last_reviewed` or mtime older than M days), and concept candidates (capitalised phrases mentioned by ≥ K notes that lack their own page). Each finding carries a path + suggestion shaped so the agent can fix via existing tools (validate_note_proposal → create_note / append_to_note / rename_note). Read-only.

v3.2.0 — Lists `.base` files (Obsidian's structured-query primitive — YAML files defining filters/views over the vault) with each base's view count and view names. Read-only. Honors `--exclude-glob` and `--read-paths`. Use this to discover which bases exist before calling `obsidian_read_base` (metadata) or `obsidian_query_base` (execute filters). Sorted by mtime descending.

Lists `.canvas` files (Obsidian's whiteboard / mind-map format — JSON nodes + edges) in the vault, with each canvas's node and edge counts. Read-only. Honors `--exclude-glob` and `--read-paths`. Use this to discover which canvases exist before calling `obsidian_read_canvas` to inspect one.

List notes in the vault. Filter by tag, folder, or modified-since date. Returns title, path, frontmatter, tags, and mtime — newest first.

Lists `.pdf` files in the vault with size + last-modified timestamp. Read-only. Honors `--exclude-glob` and `--read-paths`. Use this to discover which PDFs exist before calling `obsidian_read_pdf` to extract text. Sorted by mtime descending (newest first). PDFs are the #1 non-markdown content kind in real research vaults; this is the discovery entry point.

List every unique tag in the vault with usage counts (frontmatter vs inline). Sorted by count desc.

Close the retrieval feedback loop: after using `obsidian_search` results, call this with the note path(s) that ACTUALLY helped answer the query (pass the `path` field of the useful hits). The recorded usefulness gently boosts those notes in future searches for this vault (active only when the server was started with `--feedback-weight`). Set `useful: false` to record a note that looked relevant but was NOT helpful (lowers its boost). The per-vault cache sidecar stores the canonical absolute vault root plus relative path keys, counts, and ISO timestamps — never note content, snippets, or the qu…

Description is built at runtime.

Returns an `obsidian://open?vault=<vault>&file=<path>` URI for hand-off to the running Obsidian desktop app. No filesystem or network side effect — the URI emission lets the agent say 'open this in Obsidian' without enquire-mcp coordinating with the running app. Optional `new_pane=true` opens the note in a split. Read-only.

Walks every note for lines matching deferred-thinking markers — `Open question:` / `Q:` / `TODO?` / `??` (plus optional list-bullet/quote/heading prefixes). Returns each hit with source, the heading it lives under, line number, and age in days, sorted oldest-first so things aging out surface first. Common research-PKM pattern (Karpathy's wiki, Eleanor Konik, academic Zettelkasten). Read-only.

For each note tagged `#paper` (configurable), verify frontmatter has at least one citable identifier (arxiv / doi / url / isbn). Also flag notes whose body contains an arxiv ID (e.g. `arxiv:2401.12345`) or DOI but doesn't carry the same identifier in frontmatter — common after quick-capture from a chat. Returns each flagged note with what was found in body and a proposed frontmatter patch the agent can apply via validate_note_proposal + create_note/append_to_note. Read-only.

v3.2.0 (extended in v3.5.0) — Runs a `.base` file's filter against the vault's markdown notes, returning matching paths + the frontmatter values that contributed to the match. Supported DSL: `tag == "x"`, `taggedWith(file.file, "x")`, `linksTo(file.file, "Target")` (v3.5.0 — outbound wikilink check, basename-resolved, case-insensitive), `path startsWith "X"` / `path contains "X"` / `file.path startsWith "X"` (v3.5.0 — `file.` prefix accepted), `file.name == "X"` / `file.name != "X"` (v3.5.0 — basename equality, .md stripped), `<frontmatter_key> == <value>`, `<key> != <value>`, `<key> contains …

v3.2.0 — Parses a `.base` file into structured JSON (filters, formulas, properties, summaries, views). Does NOT execute the query — use `obsidian_query_base` for that. Useful when an agent wants to introspect the structure of a base before deciding which view to run, or to surface the base's saved queries to the user. Read-only.

Parses one `.canvas` file into typed nodes (text / file / link / group) + edges (with from/to node IDs and optional sides + labels). Each `file` node carries a `file_resolved` field — the vault-relative path that the canvas's file reference resolved to (or null if broken). The response also includes a `summary` of node-kind counts and a `broken_file_refs` array surfacing canvas files that reference non-existent notes. Read-only.

Read a note by relative path or by title (filename without .md). Default `format: "full"` returns content + frontmatter + wikilinks + embeds + tags. `format: "map"` returns just headings + frontmatter keys + counts (no body) — useful for planning a surgical edit without paying token cost for the body. Title accepts periodic-note aliases ("today"/"daily"/"weekly"/"monthly") that resolve to the standard `YYYY-MM-DD`/`YYYY-Www`/`YYYY-MM` names. Errors include `Did you mean: ...` suggestions on near-misses.

Extracts plain text from one PDF, returning per-page text, the extractor's single bounded `full_text` aggregate (not a second join), and doc-level metadata. Every page reports `status: ok | empty | failed` with bounded failure evidence when failed; `complete` is true only when every requested page is `ok` or `empty`, so `has_text: false` means a genuine blank/image-only selection only when `complete: true`. Built-in hard UTF-8 item/page/aggregate and item/node/page-result budgets are enforced during application-level extraction before retention and joins. Overflow emits terminal `PDF_TEXT_BUDG…

Atomically rename a note AND rewrite every [[wikilink]] / ![[embed]] in the rest of the vault that resolves to it — preserving |alias, #section, ^block, and the user's chosen path-qualification convention (bare basename vs path). Code-fence-aware: wikilinks inside ``` / ~~~ blocks are left verbatim. Use dry_run=true to preview which files would change without touching disk. Returns per-file rewrite counts + total. WRITE TOOL — only available when the server is started with --enable-write.

Walks the vault (or a `folder` subset), substitutes every occurrence of `search` with `replace` outside fenced code blocks (` ``` ` / `~~~`), and writes each modified file back. Reuses the same line-walker rename_note uses, so example snippets and code documentation stay verbatim. Pass `dry_run=true` to preview the plan without touching disk — you get per-file occurrence counts + total. `case_sensitive` defaults to true. Refuses identical search/replace and empty search to prevent footguns. WRITE TOOL — only registered when --enable-write is passed.

Resolve an Obsidian [[wikilink]] (or ![[embed]]) to a vault file. Handles aliases (Note|alias), sections (Note#Heading), block refs (Note^block), and ../-relative paths.

**The default search tool for v2.0.** Auto-detects every available retrieval signal — BM25 via FTS5 (if `--persistent-index`), TF-IDF cosine (always), and ML embeddings (if `enquire-mcp build-embeddings` ran) — and fuses them with Reciprocal Rank Fusion (Cormack et al, 2009) for higher recall and better paraphrase / synonym matching than any single ranker. Equal weights, k=60. Gracefully degrades: with only TF-IDF available it produces TF-IDF-style ranking; with BM25+TF-IDF it does keyword-augmented retrieval; with all 3 it provides full hybrid conceptual + lexical retrieval, free / offline / …

Case-insensitive token search across all notes. Default mode `all` requires every whitespace-separated token to appear in a note (AND-tokenizer); `any` requires at least one (OR); `phrase` does the old contiguous-substring match. Returns a structured response with `query`, `mode`, `scanned_notes`, and ranked `matches` (each with snippet, line, score, matched_terms) — empty matches are explicit, not ambiguous with a broken call.

Pure-JS weighted lexical retrieval. Tokenizes + TF-IDFs + L2-normalizes each admitted note body, then ranks by cosine similarity over exact normalized token overlap. Free, offline, and model-free: it can emphasize rare shared terms, but it does not infer synonyms, paraphrases, or cross-language meaning; use embeddings for those cases. A folder scopes both discovery and IDF. Corpus/token/byte budgets fail closed instead of returning a silent prefix, and source-generation receipts prevent stale vectors from being paired with live snippets. Returns ranked hits with snippet + matched terms (highes…

List notes not edited in N days (forgetting-aware staleness), oldest first. Use to surface facts that may be outdated before relying on them, or to pick notes to refresh. Cheap mtime-only scan; returns path / title / mtime / age_days.

Vault-wide summary: total notes, total bytes, average note length, recently-modified count (last 7 days), orphan notes (no inbound + no outbound), broken wikilink count, total tag count, and top-N tags by frequency. Cheap (one pass over the cached parse). Useful as the first call in a session so the LLM has structural context before issuing targeted reads.

Lint a draft note BEFORE writing. Closes the #1 LLM-write pain: AI generates structurally-broken notes (bad YAML, fake wikilinks, inconsistent tags). This tool parses the proposed YAML, resolves every [[wikilink]] against the live vault (broken/resolved with did-you-mean), pre-classifies every tag (existing vs new), and checks for path/title collisions. Returns errors (blocking) + warnings (non-blocking) + per-link/tag diagnostics. Always available — does NOT require --enable-write. Recommended workflow: validate → fix → obsidian_create_note.

List the frontmatter keys that exist in this vault, how many notes carry each, the value shapes observed (string / number / boolean / list / map / null), and short example values. Every other frontmatter tool needs the key as an argument, which only helps someone who already knows the vault — call this first to find out what there is to ask about. Enumeration, not ranking: exhaustive within its scan bounds, and it refuses rather than returning a partial inventory. Honors --exclude-glob / --read-paths. Read-only.