Two tools with basic schemas and descriptions, but significant gaps in parameter documentation and output schema clarity. web_search has a count parameter with a default (good), but neither tool documents output structure. Descriptions are present but generic (194 chars baseline suggests room for improvement). No error handling guidance, no parameter validation rules stated, no output schema documentation. Tools are simple and well-named (verb_noun pattern), but lack the depth expected for production use.

• Document output schemas for both tools. web_search should return: {results: [{title: string, url: string, snippet: string}], count: number}. web_fetch should return: {url: string, title: string, content: string, extracted_at: ISO8601}.
• Add parameter validation rules to descriptions. For web_search count: 'Must be 1-10; values outside this range are clamped to the nearest boundary.' For web_fetch url: 'Must be a valid HTTP/HTTPS URL.'
• Enhance error messages with recovery guidance. Instead of 'Search error: <message>', return 'Search failed: <specific reason>. Try: [1] simplify query, [2] reduce count, [3] check network.' This guides LLM retry logic.
• Clarify tool selection criteria. Add to web_fetch: 'Use this after web_search to extract full article content. Not suitable for real-time data or dynamic pages.' This prevents misuse.
• Add timeout and rate-limit documentation. State: 'web_fetch has a 10-second timeout per URL. If timeout occurs, the URL may be unreachable or very slow, try a different source.'