Production-ready MCP research server examples with logging, validation, and tooling.
Server has significant definition quality issues. Five tools are declared, but tool definitions appear inferred rather than explicitly visible in source (fastmcp registration not shown). Tool descriptions are present but minimal (8-45 chars, well below the 50-200 char LLM-optimized baseline). Input schemas are visible and include type info, but descriptions for parameters are sparse or generic. No output schemas documented. Duplicate tool names ('search' and 'fetch' each appear twice) create ambiguity. Error handling guidance is absent. The source shows only frontend/app.py and frontend/services/mcp.py snippets; core tool registration logic in fastmcp is not visible, forcing reliance on inference.
Run the coordinated evaluation and return variables.
Fetch a record by ID.
Fetch a cupcake order by ID.
Search records using the given method.
Search for cupcake orders via keyword matching.
Duplicate tool names: 'search' appears twice (records and cupcake), 'fetch' appears twice. LLMs cannot disambiguate which tool to call when names collide. Tool selection becomes unreliable.
Tool descriptions are extremely short: 'Search records using the given method.' (45 chars), 'Fetch a record by ID.' (21 chars), 'Run the coordinated evaluation and return variables.' (50 chars). Baseline for LLM-optimized descriptions is 50-200 chars. These lack context for when/why an LLM should select the tool, especially for similar-sounding tools like 'search' and 'fetch'.
Parameter descriptions are minimal or absent. 'query' is described as 'The search query string' (24 chars) and 'method' as 'The search method to use (optional)' (37 chars). No guidance on valid method values, query format, or search semantics. The 'method' parameter lacks an enum constraint, free-form strings invite hallucinated values.
Inferred effective spec: 2025-06-18+.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 46 | 2025-06-18+ | v2 |
| 2026-03-09 | F | 33 | - | v1 |
No output schemas documented. Callers (and LLMs) cannot know what fields to expect from 'search', 'fetch', or 'evaluate'. Without documented return types, LLMs cannot plan downstream operations or extract required data reliably.
No error handling guidance. Tools do not describe what errors can occur, whether they are retryable, or what the LLM should do if a call fails. A bare HTTP 400 or 404 provides no recovery path.
Tool definitions inferred from tool descriptions table rather than visible in source code. The fastmcp registration logic is not shown, forcing assessment via inference. Per HARD SCORING RULE: if tool registration is not explicitly visible, cap tool overall at 50.