This MCP server has significant definition quality gaps. While both tools have basic descriptions and input schemas, the descriptions lack depth and clarity needed for LLM selection. The 'search' tool description is entirely in Chinese, making it inaccessible to English-speaking agents. Schemas are present but lack proper enum constraints for the 'format' and 'engine' parameters despite listing discrete valid options in descriptions. Parameter descriptions are generic and lack actionable guidance on constraints, formats, or when to use each parameter. Error handling is minimal, tools return JSON error objects without recovery guidance. No output schemas are documented. Tool naming is acceptable but descriptions fall well below the 10-1024 character baseline for optimal LLM comprehension.

• Translate the search tool description to English and expand it to 100-200 characters explaining when to use search vs read_url, and which engine (DuckDuckGo vs Google) is recommended for different query types.
• Add JSON Schema enum constraints: 'format': enum=['raw_markdown', 'markdown_with_citations', 'references_markdown', 'fit_markdown', 'fit_html', 'markdown'] and 'engine': enum=['duckduckgo', 'google']. Remove enum listings from descriptions and move them to schema.
• Document output schemas explicitly. For read_url: 'Returns a string (the crawled page content in the requested format).' For search: 'Returns JSON with structure {results: [{title: string, link: string, snippet: string, source: string}, ...], error?: string}'.
• Add explicit parameter constraints: url description should state 'Valid URL starting with http:// or https://', num_results should state 'Integer between 1 and 100 (default 10)', query should state 'Non-empty search string, up to 500 characters'.
• Implement actionable error messages. Instead of {'error': 'Search error: ...'}, return '{error: string, retryable: boolean, suggestion?: string}' e.g., 'Invalid search engine. Valid options: duckduckgo, google. (retryable: false)' or 'DuckDuckGo rate limit exceeded. Retry in 60 seconds. (retryable: true)'.
• Add recovery guidance in error responses. Example: 'If URL is malformed, check it starts with http:// or https://. For 429 errors, wait before retrying. For connection timeouts, verify the URL is accessible.'
• Clarify tool differences: add to read_url description 'Use this to extract full page content. For discovering URLs, use search() first.' Add to search description 'Use this to find URLs. For extracting content from a known URL, use read_url().'