Local code intelligence and project memory for MCP coding agents, with TypeScript, JavaScript and Prisma analysis.
Server has 7 tools with minimal documentation. All tools have descriptions (10-60 chars, below the 50-200 char baseline for LLM optimization), but lack critical detail on WHEN to use each tool and what distinguishes them. Input schemas are present but sparse: search_code and search_symbols accept only a 'query' string; find_* tools accept a 'name' string; pagination tools accept 'limit' and 'offset' numbers. No output schemas documented. No error handling guidance. No parameter constraints (enums, ranges, patterns). Tools like find_callers, find_callees, and find_references are semantically similar but descriptions do not explain the distinction. No indication of idempotency, side effects, or prerequisites.
Query find_callees in the selected project
Query find_callers in the selected project
Report review candidates from the selected completed index
Report review candidates from the selected completed index
Query find_references in the selected project
Query search_code in the selected project
Query search_symbols in the selected project
Descriptions are too short (10-60 chars vs. baseline 50-200 chars). LLMs cannot determine when to select search_code vs search_symbols, or distinguish find_callers from find_callees from find_references. Descriptions must explain WHAT the tool does, WHEN to use it, and how it differs from similar tools.
No output schemas documented. LLMs cannot plan downstream tool calls or extract required fields (e.g., does search_code return file paths, line numbers, code snippets?). Every tool must document its return type and structure.
Input schemas lack constraints. 'query' and 'name' parameters have no length limits, regex patterns, or format guidance. 'limit' and 'offset' have no min/max bounds (e.g., is limit 1 - 1000? Is offset 0-based?). Unbounded parameters invite invalid LLM inputs.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
No error handling guidance. If a search returns no results, or a symbol is not found, what does the tool return? How should the LLM recover? Error responses must guide the agent on next steps (retry, try a different query, etc.).
Semantic overlap not clarified. find_callers, find_callees, and find_references are distinct operations but descriptions do not explain when to use each. LLMs will guess and waste calls. Descriptions must state: 'find_callers: returns functions that call the given symbol' vs 'find_callees: returns functions called by the given symbol' vs 'find_references: returns all code locations that reference the symbol (including non-function contexts)'.