The server exposes a single semantic code search tool with well-structured input schema and comprehensive parameter documentation. The tool description is clear and action-oriented (127 chars), explaining WHAT the tool does (semantic search), WHEN to use it (instead of grep/glob), and provides concrete examples. All five input parameters have type definitions and descriptions. However, the output schema is not documented in the source code provided, the response structure is not visible, which prevents verification of return field quality, pagination patterns, or chaining capability. The tool name 'search' is generic but contextually clear given the server's purpose (cocoindex-code). Parameter defaults are sensible (limit=5, offset=0, refresh_index=true), and constraints are properly specified (limit 1-100). No security or error handling implementation visible in the provided source snippet.

• Document the output schema explicitly: e.g., 'Returns a SearchResult object with fields: code_snippet (string), file_path (string), line_number (integer), relevance_score (float 0-1), language (string).'
• Add pagination guidance: 'The response includes total_count (total matches in codebase) and has_more (boolean). Use offset + limit to paginate through all results. Start with limit=5 for quick iteration; increase if most results are relevant.'
• Include error scenarios: 'If no index exists for the codebase, returns error code INDEX_NOT_FOUND with message "Codebase index not found. Run 'ccc index' to build it first." If query times out after 30s, returns TIMEOUT error, retry with a more specific query or smaller codebase scope.'
• Consider renaming to 'search_code' or 'semantic_code_search' to reduce naming ambiguity in multi-tool compositions.
• Add parameter dependency hint: 'If you only want to search Python files, use languages=["python"]. To search src/ and lib/, use paths=["src/*", "lib/*"].'
• Document refresh_index behavior: 'Set refresh_index=true (default) for initial searches or after code changes. For repeated queries on an unchanged codebase, set refresh_index=false to skip the ~X ms indexing overhead and return cached results.'
• Clarify what 'natural language query' means: 'Queries can be feature descriptions (e.g., "user authentication"), implementation patterns (e.g., "error handling"), or code snippets. The index uses semantic embeddings to find conceptually similar code, not keyword matching.'