Self-contained hybrid (BM25 + static-embedding) AST code-search MCP server
Three tools with clear names and documented schemas. search and status have good descriptions (90+ chars); set_roots is minimal (47 chars). All tools have input schemas with type definitions and parameter descriptions. However, output schemas are only partially documented in code comments (StatusResult, CodeChunkResult, SearchResultModel are Pydantic models but not explicitly linked in tool registration). Error handling is absent, no guidance on what happens if indexing fails, roots are invalid, or search returns no results. No tool annotations (readOnlyHint/destructiveHint) despite set_roots being a WRITE operation. Parameter descriptions are adequate but lack format constraints (e.g., 'roots' array has no min/max length, no path validation hints).
hybrid code search across all resolved roots
override the indexed workspace directories
indexing status, chunk/file counts, and language breakdown
set_roots description is only 47 characters ('override the indexed workspace directories'), below the 50-char minimum for clarity. Does not explain consequences (rebuilds index?), prerequisites (absolute paths required?), or error cases.
No tool annotations (readOnlyHint/destructiveHint/idempotentHint) despite set_roots being a WRITE operation that modifies indexed roots. LLMs cannot infer side effects without explicit hints.
Error handling is absent. No guidance on recovery if indexing fails, roots are invalid, or search times out. SearchResultModel has a 'message' field but no documented error codes or retry logic.
Parameter 'roots' in set_roots lacks constraints: no min/max array length, no path validation hints (absolute vs relative), no guidance on what happens if a path is invalid or inaccessible.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | C | 67 | 2026-07-28+ | v2 |
Output schemas (StatusResult, CodeChunkResult, SearchResultModel) are defined as Pydantic models in code but not explicitly documented in tool registration. LLMs must infer structure from code comments rather than explicit schema declarations.