WebCat has 5 tools with mixed quality. Tool names follow verb_noun convention (search, scrape, get) which is good. However, descriptions vary significantly in completeness. The 'deep_research' tool has a strong description (160+ chars, explains timing and effort levels), while 'health_check' and 'get_server_info' are minimal (12 and 9 chars respectively). Input schemas are present for all tools but lack depth, most parameters have minimal or generic descriptions. Output schemas are not documented in the source provided. Error handling and recovery guidance are absent. The server relies on fastmcp framework which likely generates schemas, but explicit parameter descriptions for most tools are minimal or missing entirely.

• Expand 'health_check' description to ~100 chars: 'Check if the WebCat server is running and its dependencies (Serper API, DuckDuckGo, Perplexity AI) are accessible. Returns status and uptime.'
• Expand 'get_server_info' description: 'Retrieve configuration, available APIs, model versions, and capability flags. Call this first if unsure what the server supports.'
• Document output schemas for all tools. E.g., 'search' returns: {results: [{title: string, url: string, snippet: string, source: string}], total_count: int, query: string}
• Add parameter constraints: 'max_results' should specify 'minimum: 1, maximum: 50' and explain cost/latency tradeoffs.
• Add error recovery guidance to 'search' and 'deep_research': 'If Serper API quota exceeded, try DuckDuckGo fallback. If Perplexity API fails, fall back to search + scrape_url for manual research.'
• Clarify 'research_effort' enum: 'low (30-60 sec, ~10 sources), medium (2 min, ~30 sources), high (2-4 min, ~100 sources). High effort costs more API quota.'
• Document pagination: if 'search' returns max_results=5 but total_count=50, add next_cursor field and explain how to fetch page 2.
• Add dependency hints: 'search_urls should be called before scrape_url if the agent doesn't already have URLs.'
• Validate inputs: 'max_results' should reject negative or >100 values with actionable error: 'max_results must be 1-50; got {value}.'
• Consider splitting 'deep_research' into two tools: 'deep_research_async' (returns job_id) and 'get_research_result' (polls job_id), since 2-4 minute latency blocks the agent. Alternatively, document timeout behavior.

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

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