Multi-vault Obsidian filesystem-first MCP server with vault routing, SQLite FTS5 indexing, and SPARK method PKM support
The Obsidian MCP server has 13 tools with basic JSON Schema definitions and descriptions. However, the quality is inconsistent and several critical issues prevent a higher score. Tool naming follows verb_noun convention correctly (obs_search_notes, obs_read_note, etc.). Descriptions are present for all tools but are often generic and lack detail about WHEN to use each tool or what distinguishes it from similar tools (e.g., obs_search_notes vs obs_keyword_search lack clear differentiation). Parameter descriptions are minimal and do not adequately document constraints, formats, or valid ranges. No output schemas are documented, it's unclear what fields to expect from responses. Error handling guidance is absent. The server supports destructive operations (obs_delete_note) without confirmation patterns. Security risks are present: no evidence of permission gates or audit logging visible in source code.
Append text or a new section to an existing note
Create a new note with optional frontmatter and SPARK entity/event validation
Delete a note from the vault
Build a rich dossier for an entity (contact/account) with connections, events, backlinks, and engagement metrics
Audit vault for schema violations, unresolved links, alias collisions, and missing required frontmatter in entity cards
List journal/daily notes within a date range in a specific scope
List notes in a scope (folder) with optional tags, time filtering, and pagination
No output schemas documented. Tools return structured responses but agents cannot know what fields to expect (required for tool chaining and downstream field extraction). obs_search_notes, obs_list_notes, obs_get_vault_structure return metadata and content but this is undocumented.
Destructive tool (obs_delete_note) lacks confirmation/dry-run pattern. Agents can irreversibly delete notes without explicit confirmation step. No warning in description that this is destructive.
obs_search_notes and obs_keyword_search both search note content. Descriptions fail to clarify when to use one vs the other. obs_search_notes says 'full-text search' but obs_keyword_search says 'search for keyword', the distinction is unclear to LLMs. See naming pattern for avoiding conflicting tool names.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 11 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 39 | 2024-11-05+ | v1 |
Check if a note exists at a given path
Health check endpoint for the MCP server
Query notes by frontmatter fields with filtering operators (gte, lte, gt, lt, eq, ne)
Read the full content and frontmatter of a note at a given path
Resolve a bare entity name or wikilink to its full path and parsed metadata
Search notes by keyword with optional folder/scope filtering and case sensitivity
Update an existing note's content and optional frontmatter, with entity schema validation
Get the folder and file structure of a vault, optionally cached
List available workspaces/vaults and their metadata
Parameter descriptions are minimal and do not state constraints, formats, or valid ranges. E.g., 'folder' param in obs_search_notes lacks guidance on expected format (relative path? absolute path? with/without leading slash?). obs_execute_command accepts arbitrary command IDs with no enum or list of valid commands. See pattern:constrained-input.
obs_execute_command has no validation or documented list of valid Obsidian command IDs. Agents can pass arbitrary strings, risking invalid commands. Should document expected format, provide examples, or ideally enum valid commands.
No error handling guidance. Descriptions do not explain what errors can occur or what agents should do if a tool fails. E.g., obs_read_note does not document behavior when path does not exist. See pattern:recovery-guide.
Parameter 'path' used across tools but not consistently documented. obs_read_note expects 'folder/note.md' format (from description), but ob_create_note has same param without format guidance. Should specify: relative vs absolute, required extension, encoding, max length.
obs_get_vault_structure 'use_cache' boolean parameter has no explanation of what cache is, when to use it, or what impact caching has on result freshness. Should state default behavior and explain tradeoff.
obs_keyword_search has 'limit' parameter defaulting to 50 but no documentation of what happens if results exceed limit (are they truncated? paginated?). obs_list_notes lacks pagination/limit parameters entirely, risking context window exhaustion on large vaults. See pattern:paginated-result.
No evidence of permission gates or audit logging in source code. Tools like obs_delete_note and obs_execute_command can modify vault state with no documented access control. Agents need to know what permissions are required.