Strong tool naming (all verb-prefixed, clear intent) and comprehensive descriptions (avg 180 chars, well above 10-char minimum). All 13 tools have explicit schemas with proper Zod types and parameter descriptions. Input validation is present via Zod parsing. However, output schemas are not formally documented in the tool definitions, responses are text-based without structured field declarations. Error handling is minimal; tools do not guide recovery or categorize errors. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite all being read-only operations. Parameter descriptions are good but could be more prescriptive about constraints (e.g., search query minimum 4 chars is mentioned in description but not enforced in schema).

• Add formal output schemas to each tool definition. Define structured return types (e.g., { principles: [{ id: string, title: string, description: string, guidelineCount: number }] }) so LLMs can extract and chain data.
• Add readOnlyHint: true to all 13 tool definitions to signal idempotency and retry safety to agents.
• Implement input validation with clear error messages. For search-atag, enforce minLength: 4 in the Zod schema and return 'Search query must be at least 4 characters' on violation.
• Add pagination to list tools: accept limit (default 20, max 100) and offset parameters; return total count and next_offset in response. Cap results at 50 items per call.
• Document what each tool returns in its description. E.g., 'Returns a list of principles with ID, title, description, and guideline count.' This helps LLMs understand what data is available for downstream use.