MCP server for Stash quick-capture panel. Reads the Stash data file directly (read-only) and writes new captures to a sidecar inbox file. Provides tools to list, search, and add items to the user's local Stash.
Three tools with clear verb-noun naming (list_items, search_items, add_item). All have descriptions (194 - 250 chars, within baseline 34 - 392). Input schemas present with Zod validation and descriptions. Tool annotations correctly applied (readOnlyHint for queries, destructiveHint=false for write). Key gaps: no output schema documentation, no pagination support for list_items/search_items, error handling lacks recovery guidance, no parameter constraints (e.g., section name format, text length limits beyond minLength=1).
Capture a new item into the user's Stash inbox. It appears in the app within a few seconds (top of the list, or in the named section if one matches).
List captured items from the user's local Stash quick-capture panel. Optionally filter by section name or pending (not done) only.
Case-insensitive substring search across all captured Stash items.
No output schema documentation. Callers cannot predict response structure (text content, field names, pagination). LLMs must infer from examples.
list_items and search_items lack pagination (limit, offset, cursor). Large result sets will blow context window; no guidance on result limits.
Error responses are generic (catch-all returns empty list or no guidance). Missing recovery hints: 'Section not found. Available sections: [...]' or 'Search returned no results. Try a shorter query.'
add_item accepts free-form section name with no validation. If user passes 'My Section' but actual section is 'my-section', silent mismatch occurs. No enum or format constraint.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 62 | 2026-07-28+ | v2 |
add_item description does not clarify idempotency. If called twice with identical text, does it create duplicates or deduplicate? Agents need to know if retry is safe.