The server provides 15 tools with explicit schemas and descriptions in src/services/tool-service.ts. Naming follows verb-first convention (search_, filter_, get_, etc.), but descriptions are in Chinese and vary in quality. Schemas are present and mostly complete with type definitions and required field declarations. However, several critical gaps reduce overall quality: (1) Descriptions are minimal (10-50 chars), well below the 50-200 char LLM-optimized baseline. (2) Many parameter descriptions lack actionable guidance on constraints and valid ranges. (3) Output schemas are not documented, tools return ToolResponse objects but the structure is not specified. (4) No error recovery guidance is present. (5) The sync_repository_data tool is marked WRITE but accepts minimal parameters, and check_new_apis returns new API records but lacks pagination support despite potentially large result sets. The tools form a coherent API discovery domain but fall short of production-grade quality.

• Expand all tool descriptions to 50-200 characters. Follow the pattern: WHAT does it do? WHEN should I use it? WHAT does it return? E.g., 'search_apis_by_keyword': 'Search public APIs by keyword. Returns matching APIs with names, base URLs, and authentication requirements. Use to find APIs by functionality (e.g., "weather", "payment").'
• Document output schemas for every tool. Define the ToolResponse structure: which fields are returned, their types, and how pagination is handled. Example: '{"content": [{"type": "text", "text": "..."} ], "items": [ { "name": string, "baseUrl": string, "auth": string, "https": boolean } ], "totalCount": number }'
• Add constraint descriptions to parameters. For 'limit', specify: 'Number of results to return (1-100, default 10)'. For 'projectType', provide: 'Short description of the project domain (e.g., "weather app", "payment processor", "social media").'
• Implement pagination for list-returning tools. Add 'offset' and 'limit' parameters, or 'cursor' + 'limit'. Return 'totalCount', 'hasMore', and 'nextCursor' in responses. Cap default limit at 20-50 to prevent context explosion.
• Translate all descriptions to English and provide English descriptions alongside or instead of Chinese. LLMs operate primarily in English; Chinese descriptions add translation overhead and error risk.
• Add error recovery guidance to the handleToolCall() method documentation. Specify: (1) What errors can occur (GitHub rate limit, API not found, etc.). (2) Which are retryable. (3) What the LLM should do (wait, search by different criteria, etc.).