CodeVault has 8 well-named tools with complete input schemas and descriptions. Tool names follow verb_noun convention (search_code, index_project, ask_codebase). All tools have descriptions (avg ~60 chars, within baseline). Input schemas are properly typed with Zod validation. However, output schemas are not documented, LLMs cannot predict response structure. Parameter descriptions are minimal (e.g., 'Search query' lacks format/constraint guidance). Error handling is structured (MCPErrorPayload with code/type/message) but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Missing: pagination guidance for search results, idempotency guarantees, and per-parameter constraint documentation.

• Document output schemas for all tools. E.g., search_code returns [{sha, path, score, language, chunk_text}]; ask_codebase returns {answer, citations: [{sha, path, line_range}]}. Use JSON Schema in tool definition or detailed description.
• Add min/max bounds to numeric parameters: limit (1 - 200), max_chunks (1 - 100), temperature (0 - 2 already present). Add enum descriptions: provider 'auto' = auto-detect, 'openai' = use OpenAI API.
• Add tool annotations: search_code/get_code_chunk/get_project_stats → readOnlyHint: true; index_project/update_project/use_context_pack → destructiveHint: true (or idempotentHint if safe to retry).
• Enhance error suggestions. E.g., ENCRYPTION_KEY_REQUIRED → 'Set CODEVAULT_ENCRYPTION_KEY environment variable'; PATH_VALIDATION_FAILED → 'Ensure path is inside project root or use default "."'.
• Document pagination for search results. E.g., 'Returns up to limit results. For large codebases, use path_glob or tags to narrow scope. No cursor pagination yet, results are unordered.'

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

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