Two tools with basic schemas and minimal descriptions. search-web has a 54-char description (below baseline 194 chars); extract-info-from-search has 60 chars. Both schemas present but lack output documentation. No error handling guidance, no parameter constraints (e.g., page_number bounds), no idempotency hints. Tool names are verb-based (search-, extract-) but descriptions are vague about what data is returned or when to use each tool. No indication of whether tools modify state or are read-only (though marked READ_ONLY in metadata). Missing parameter descriptions for page_number (what range? what happens if out of bounds?). No guidance on pagination limits or result structure.

• Expand search-web description to 150 - 250 chars: 'Search the web for information and scrape results. Use this to find current data, news, or answers to user questions. Returns a list of results with title, URL, and snippet. Supports pagination via page_number (1-based, default 1).'
• Expand extract-info-from-search description: 'Extract and merge relevant information from previous search results. Call this after search-web to synthesize findings across multiple queries into a cohesive summary. Requires prior search-web calls in the conversation.'
• Add output schema documentation in tool definitions: specify that search-web returns {results: [{title, url, snippet, rank}], total_count, page, has_next_page} and extract-info-from-search returns {summary: string, key_points: [string], sources: [url]}.
• Add bounds and description to page_number: 'Page number for pagination (integer, 1-based, default 1). Maximum 100 pages. Returns error if page exceeds available results.'
• Implement error handling in call_tool handlers: catch exceptions from perform_search_and_scrape and find_similar_chunks, return structured errors with recovery hints (e.g., 'No results found. Try broader keywords or check internet connection.').

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry