Two tools with complete input schemas and clear descriptions. web_search and web_extractor both have well-structured parameters with types, descriptions, and sensible defaults. Descriptions are concise (90-120 chars) and action-oriented. However, output schemas are not documented in the visible code, only inferred from implementation. Parameter descriptions could be more prescriptive about constraints (e.g., exact enum values for 'mode', 'categories'). Error handling is mentioned in descriptions ('never raises') but recovery guidance is absent. No tool annotations (readOnlyHint, destructiveHint) despite both being read-only operations.

• Document output schemas explicitly. For web_search, specify: {query: string, results: [{title: string, url: string, snippet: string, ...}], error?: string}. For web_extractor, specify: {results: [{url: string, markdown: string, word_count: int, status: string, error?: string}], error?: string}.
• Convert 'categories' and 'mode' parameters to enums in JSON Schema. E.g., categories: enum=['general', 'news', 'science', 'it', 'images']; mode: enum=['fit', 'raw', 'bm25', 'llm']. Remove example text from descriptions.
• Add readOnlyHint: true to both tool schemas to signal they are safe to retry and do not modify state.
• Enhance error descriptions with recovery guidance. E.g., 'If provider fails, the tool returns {error: "<reason>"}. Retry with a different provider parameter, or ask the user to refine the query.'
• Clarify time_range null behavior: 'null (default) = no time filter; backend applies its default recency window.'