Two tools with complete input schemas and reasonable descriptions. tavily-search has a comprehensive parameter set with enums and constraints (search_depth, topic, time_range, country, max_results bounds). tavily-extract is simpler but well-defined. Descriptions are adequate (194 - 250 chars, within baseline 34 - 392 range) but lack actionable guidance on when to use each tool vs. the other, or what downstream steps follow. No output schema documentation visible. Error handling is minimal, no recovery guidance or categorization. Parameters are well-typed with enums where appropriate, but descriptions could be more prescriptive about format/constraints (e.g., 'country' enum is massive and could benefit from a note on how to handle user input like 'USA' vs 'united states'). No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite both being read-only operations.

• Document output schemas for both tools. Specify that tavily-search returns {query, results: [{title, url, content, score, published_date?, raw_content?, favicon?}], images?, follow_up_questions?} and tavily-extract returns {base_url, results: [{url, raw_content, favicon?}], response_time}.
• Add readOnlyHint: true to both tool definitions in setupToolHandlers(). This signals to agents that these calls are safe to retry.
• Enhance descriptions with recovery hints: 'Use tavily-search for broad queries; use tavily-extract to get full page content from specific URLs. If search returns no results, try a simpler query or different time_range.'
• Reduce country enum to top 20 - 30 most common countries, or accept country names/codes and validate server-side. Document: 'Country name (e.g., "United States", "Canada") or ISO 3166-1 alpha-2 code (e.g., "US", "CA").'
• Wrap API calls in try-catch with actionable error messages. E.g., 'API key invalid. Check TAVILY_API_KEY environment variable.' or 'Rate limit exceeded. Wait 60 seconds before retrying.' Return {error, retryable: boolean, guidance: string}.