Read-only MCP server over an Obsidian vault: ranked search, note reading and full-text grep.
Four read-only tools with complete input schemas and descriptions. Naming follows verb_noun convention (vault_search, vault_get_note, vault_grep, vault_list_topics). All tools have 50-200 char descriptions explaining WHAT and WHEN. Parameters are typed with constraints (min/max length, enums for response_format). Output schemas are not explicitly documented in code, responses are inferred from implementation. Error handling is minimal; no recovery guidance. Security is strong (read-only, path traversal guards). Composition is clean: each tool has one responsibility, and search→get_note forms a natural chain.
Read one note's full markdown.
Full-text search across note bodies.
Enumerate topics with note counts.
Search the vault's note manifest — START HERE for any vault question. Searches vault-index.json, a flat manifest carrying every note's topic, tags, and one-line summary. This is the fastest way to find the right note. Follow up with vault_get_note on the best path to read it in full.
Output schemas not documented. Tool descriptions state WHAT is returned (e.g. 'Read one note's full markdown') but not the structure. LLMs cannot plan downstream operations without knowing response fields.
Error handling lacks recovery guidance. No tool description explains what to do if a note is not found, a regex fails, or the vault is misconfigured. Errors should guide the LLM to retry or try an alternative tool.
vault_list_topics description is vague ('Enumerate topics with note counts'). Does not explain when to call it or what structure it returns. Should state: 'Call this first to discover available topics before searching.'
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 61 | <=2025-11-25 | v2 |
vault_grep regex parameter lacks format guidance. Description says 'Treat query as a regular expression' but does not specify Python regex syntax, escaping rules, or common pitfalls. LLMs may pass invalid patterns.
No pagination or result-limiting guidance for vault_grep. If a query matches 1000+ lines, the response could exhaust context. Description should state the limit (100 lines max) and suggest narrowing the query if truncated.