fp-mcp implements a single tool (create_note) with a proper JSON Schema and basic description. However, the server exhibits critical issues: (1) the create_note tool is directly visible in code with explicit schema registration (ListToolsRequestSchema handler), so it is NOT inferred; (2) the description 'Create a new note' is only 18 characters, below the 20-char hard floor for descriptions; (3) parameter descriptions ('Title of the note', 'Text content of the note') are minimal at 15-21 chars each, providing no guidance on constraints, format, or edge cases; (4) no output schema is documented, the response structure is hard-coded but never formally specified; (5) error handling is minimal ('Title and content are required' is the only validation check); (6) the server uses resources and prompts (good for feature richness) but the single tool lacks production maturity. The code reveals a non-functional ReadResourceRequestSchema handler that references an undefined 'notes' object (line 'const note = notes[id]', notes is never defined; the database is 'notesDb'), which would crash at runtime. This suggests incomplete testing and maintenance.
Create a new note
Tool description is too short (18 chars). Must be 10 - 1024 chars; scores below 20 chars forced to ≤40.
Parameter descriptions lack actionable guidance. 'Title of the note' and 'Text content of the note' do not specify format, length, character restrictions, or examples of valid vs invalid inputs.
Output schema is not documented. The tool returns {content: [{type: 'text', text: '...'}]} but this structure is never formally specified in a schema or description. LLMs cannot plan downstream steps or validate the response structure.
Runtime error in ReadResourceRequestSchema handler: references undefined 'notes' object. Code uses 'const note = notes[id]' but the database is 'notesDb' (Fireproof instance). The create_note tool also populates 'notes' object which is never initialized, causing a crash on first read.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 41 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 38 | - | v1 |
No idempotency guarantee. create_note generates an ID via 'String(Object.keys(notes).length + 1)' which is not idempotent, calling it twice with the same arguments produces different IDs each time. Agents retrying on transient failures will create duplicates.
Minimal error handling and no recovery guidance. The only validation is checking if title and content are falsy. No guidance on maximum lengths, special characters, encoding, or how to handle failures. Error message 'Title and content are required' does not tell the LLM what action to take next.