Loading ToolBench
Your Obsidian vault, connected to AI. MCP server for bridging personal knowledge with LLMs.
MindCache has 44 tools with inconsistent definition quality. Tool names follow verb_noun convention well (search, find_related, create_note, delete_note). However, most tools lack visible input schemas in the provided source, only 6 tools (search, ask, find_related, find_by_tag, find_by_date, add_link, suggest_connections) show explicit parameter definitions. The remaining 38 tools have descriptions in the Astro component but NO visible schema definitions in the server code. Descriptions vary widely: some are detailed (search: 150+ chars with operator examples), others are minimal (find_recent, read_section, reflect: <20 chars). No output schemas are documented. Error handling and recovery guidance are absent. Tool definitions appear inferred from the website component rather than explicitly registered in server.ts.
Add a [[wikilink]] to a note, connecting it to another note in the vault. Appends the link under a "Related" heading or at the end of the note.
Add tags to a note
Append to existing note
Ask a natural language question about your vault. Searches for relevant notes and returns their content so you can synthesize an answer. Use this when you need to understand what the user knows about a topic.
Add or remove tags across multiple notes
Mark task as done
Create note with optional template
38 of 44 tools lack visible input schemas in server code. Only 6 tools (search, ask, find_related, find_by_tag, find_by_date, add_link) show explicit parameter definitions. Remaining tools have descriptions in Astro component but no schema registration visible in server.ts.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 39 | <=2025-11-25 | v2 |
List all daily notes with stats
Permanently delete a note
Export as clean markdown or JSON
All wikilinks and URLs from a note
Find notes created or modified within a date range. Supports natural language like "today", "this week", "last month", "last 7 days", or explicit "YYYY-MM-DD to YYYY-MM-DD".
Find all notes with specific tags. Supports multiple tags (all must match). Use without # prefix.
Find broken links in the vault — [[wikilinks]] that point to notes that don't exist. These represent knowledge gaps or notes waiting to be written.
Where a term is mentioned
Find orphan notes — notes with no incoming or outgoing links. These are disconnected from the knowledge graph and might need connecting.
Recently modified notes
Find notes related to a given note via backlinks, shared tags, or shared wikilinks. Returns notes that are connected in the knowledge graph.
Find checkbox tasks
Folder tree
All tags with counts
Append to daily note
Add task to daily note
Move to different folder
Full note content
YAML frontmatter only
Content under a specific heading
Note with metadata
Daily notes from current week
Today's daily note
Reflection prompts
Quick capture to inbox
ADR-format decision log
Capture an idea
Save something learned
Structured meeting notes
Person or contact note
Bookmark with summary
Rename a note
Full-text search across your Obsidian vault. Supports operators: "exact phrase" for exact matching, path:folder to filter by path, tag:#tag to filter by tag, file:name to filter by filename. Returns matching notes with context snippets.
Analyze a note and suggest other notes that could be linked to it. Uses shared tags, mentions, and content similarity to find connections.
Modify frontmatter
Replace section content
Stats, folders, tags
38 tools have descriptions under 20 characters (e.g., 'Recently modified notes', 'Full note content', 'Reflection prompts'). LLMs cannot determine when/why to select these tools.
No output schemas documented for any tool. LLMs cannot plan downstream calls or extract required fields. E.g., does search() return note_id, path, content, and snippet? Does find_tasks() return task_id, status, due_date?
No error handling or recovery guidance visible. Tools like delete_note (destructive) lack confirmation/dry-run patterns. No error messages guide LLM on retryability, user-fixable issues, or next steps.
Tool definitions inferred from Astro website component, not explicitly visible in server.ts registration. Actual tool registration code (registerFindTools, registerReadTools, etc.) not provided in source excerpt.