A local markdown document management and search MCP server with FAQ and metadata support
NextFlows has 7 tools with basic schemas and descriptions, but significant gaps prevent production use. Tool names follow verb_noun convention (list_, read_, search_, add_, delete_, get_). Descriptions exist but are generic (10-60 chars, well below the 194-char baseline). Input schemas use Zod with type definitions, but parameter descriptions are minimal or missing. The schemas file defines richer constraints (min/max length, limits) but these are NOT used in the actual tool registrations in index.ts, a critical disconnect. Error handling is minimal (generic 'Failed to...' messages). No output schemas documented. No pagination, no idempotency hints, no permission gates. The server is STDIO-only, capping protocol readiness at 50.
Create and save a new markdown note in the data directory.
Delete an existing markdown note from the data directory.
Retrieve an answer from FAQ
Retrieve metadata for a document
List all available markdown documents in the local data directory.
Read the full contents of a specific markdown document.
Searches local markdown files for specific keywords or phrases.
Descriptions are too short (10-60 chars vs 194-char baseline). Generic phrases like 'Retrieve an answer from FAQ' lack context for LLM tool selection. No guidance on when to use get_faq vs get_metadata.
Schemas defined in src/schemas/index.ts (with min/max, limits, detailed descriptions) are NOT imported or used in index.ts tool registrations. Actual registrations use bare Zod objects. This disconnect means constraints are dead code.
No output schemas documented. Callers cannot predict response structure. Tools return JSON strings wrapped in text content blocks, no typed field definitions.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | D | 50 | 2026-07-28+ | v2 |
Error messages are generic ('Failed to read directory', 'Search failed'). No recovery guidance. LLM cannot determine if error is retryable, user-fixable, or fatal.
delete_note is destructive but has no confirmation step, dry-run option, or permission gate. Agents can permanently delete files without safeguards.