MCP server for semantic code search and retrieval using vector embeddings and ChromaDB
All three tools have input schemas with proper types and descriptions in Chinese. However, descriptions are minimal (20-40 chars) and lack critical context about when to use each tool, what they return, and how they differ. Parameter descriptions are present but terse. No output schema is documented, SearchResult dataclass exists in code but is not exposed in tool definitions. No error handling guidance. Tools lack English descriptions, limiting accessibility. Schema quality is moderate but output documentation is absent.
使用核心关键词进行初始搜索。当发现相关关键词时,也要调用此方法,而不是从本地搜索。
基于文件路径或类名搜索相关引用。发现相关文件时,不要从本地去查找相关文件,而是调用此方法
在关键词搜索中发现相关上下文后,使用此方法对搜索问题和上下文信息进行补充搜索
Output schema not documented. SearchResult dataclass exists in code but tool definitions do not declare what fields are returned (code, metadata, similarity_score, references). LLMs cannot plan downstream operations without knowing response structure.
Descriptions are under 50 characters and lack context. 'Use core keywords for initial search' does not explain when to call this vs search_by_reference, what similarity_score means, or how results are ranked. Descriptions should be 50-200 chars with WHEN/WHY guidance.
No error handling guidance. Tools do not document what happens if query_embedding fails, collection.query() returns empty results, or ChromaDB is unavailable. LLMs have no recovery path.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 45 | 2026-07-28+ | v2 |
Descriptions are in Chinese only. English descriptions are required for international LLM compatibility and code review. Non-English tool definitions reduce discoverability and maintainability.
Parameter 'n_results' lacks constraints. No min/max documented. LLMs could pass 0, 1000000, or negative values. Should specify range (e.g. 1-100) in description and schema.