This server has three tools with basic input schemas and descriptions, but significant quality gaps prevent it from reaching production-grade status. All three tools accept project_path as a string parameter (present with description), but lack comprehensive parameter validation, output schema documentation, and error handling guidance. Tool descriptions are present but generic (10-100 chars range). The 'query' tool has a numeric parameter (top_k) with a default, which is good, but lacks min/max bounds. No tool descriptions explain what happens on error or what the return structure contains. Critical patterns missing: structured output documentation, error recovery guidance, idempotency clarification, and permission gates. The codebase shows background threading for indexing, but no indication of how failures are handled or reported to the LLM. Overall design treats tools as simple wrappers around Python methods without LLM-specific guidance.

• Document output schemas for all tools. E.g., full_index should return {status: 'ok'|'error', message: string, task_id?: string}. status should return {last_index_time: ISO8601, total_files: int, file_hashes: {[filename]: hash}, metadata_path: string}. query should return {results: [{file: string, line_range: [int, int], score: float, code_snippet: string}], total_count: int, next_cursor?: string}.
• Add min/max constraints to numeric parameters. top_k should be 1 - 100 with default 10. Add length limits to text queries (max 2000 chars).
• Replace bare error handling with typed error responses. Categorize: (1) retryable (e.g., 'Index still building, retry in 30s'), (2) user-fixable (e.g., 'Project path not found, provide absolute path'), (3) fatal (e.g., 'OpenAI API key not set'). Provide recovery guidance.
• Clarify idempotency and side effects in full_index description. E.g., 'Idempotent. If indexing is in progress, returns immediately. If completed, skips re-indexing. Spawns daemon threads for periodic updates. Call once per project; repeated calls do not re-index.'
• Add a discovery tool (list_indexed_projects) to help LLMs find valid project paths without trial-and-error.
• Implement permission checks. Verify caller (via agent context) has read/write access to project_path before indexing or querying. Log all calls with caller, timestamp, project, query text (redacted if sensitive), and result count.
• Add pagination to query tool. Support offset/limit or cursor-based pagination. Include total_count in response so LLM knows if more results exist.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry