A FastAPI-based server providing query handling against simulated long-term memory with agent knowledge base
This MCP server has a single tool 'handle_query' with critical gaps in schema definition, parameter documentation, and error handling. While the tool name starts with a verb, the schema is incomplete, parameters lack descriptions, and the implementation does not follow production-grade tool patterns. The tool description is present but minimal (68 chars, below the 194 char baseline). Input schema exists but lacks critical parameter-level descriptions. Output schema is undocumented. No error classification, no recovery guidance, no pagination, no result limits. The server is a bare proof-of-concept, not production-ready.
Handles query requests against agent memory, searching for matching keys and returning associated memory values
Input schema present but parameter 'query' lacks description in the schema; LLM cannot infer intent or constraints from name alone
No output schema documented; LLM cannot predict response structure or plan downstream tool calls
Tool description is only 68 characters; baseline is 194 chars (p10=34, p90=392). Too brief to guide LLM selection or explain when to use this tool vs alternatives
No error handling or recovery guidance; returns generic 'No memory found' without explaining what to do next or how to refine the query
Query matching is case-insensitive substring search; no documentation of this behavior, no parameter constraints (e.g., min/max length), no guidance on expected format
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 42 | 2026-07-28+ | v2 |
| 2026-03-09 | F | 33 | - | v1 |
Tool does not declare its scope or permissions; no audit trail capability; memory is simulated and hardcoded, not real external state
No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite being READ_ONLY; misses explicit signal to LLM that this tool is safe to retry