Server defines 3 web tools with clear action-verb naming (search_web, fetch_url, view_website) and comprehensive parameter schemas using Pydantic Field annotations. All tools have descriptions (10 - 200 chars range, solid) and input schemas with type constraints. However, output schemas are undocumented, callers cannot predict the structure of returned data. Error handling is minimal (no recovery guidance). Tool compositions are sound (each does one thing), but lacking inter-tool chaining hints. Descriptions are adequately detailed but do not reference dependencies or failure modes. The 'limit' parameter in fetch_url (hardcoded to 20,000 in implementation) is not reflected in the schema, creating a discrepancy.

• Add output schema documentation for all three tools. For search_web, specify: {results: [{title: str, url: str, snippet: str, ...}], total_results: int}. For fetch_url, specify: {content: str, content_type: str, metadata: {...}}. For view_website, specify: {format: 'png'|'jpeg', data: bytes, width: int, height: int}.
• Expose the 'limit' parameter in fetch_url schema with description 'Maximum characters to return (default 20000, max 100000).' Update the implementation to accept this parameter instead of hardcoding.
• Add error handling examples to each tool. For search_web: 'If all providers are unavailable, the tool will return an error with available fallback providers.' For fetch_url: 'If the URL returns a 4xx/5xx error, returns {error: str, status: int, suggestion: str}.' For view_website: 'If rendering times out after 10s, returns {error: "Timeout", suggestion: "Try fetch_url for text-only content"}.'
• Enhance view_website description: 'Captures a rendered screenshot of a webpage, useful when visual layout or interactive elements matter more than text. Returns a PNG image. Use fetch_url for faster text extraction; use view_website when you need to see the visual design, detect broken layouts, or verify form fields are visible.'
• Document fetch_provider tradeoff: 'trafilatura: fastest, works for most articles; httpx: fallback for JavaScript-heavy sites; zendriver: handles dynamic rendering but slower. Auto selects based on heuristics.'
• Add tool-chaining hints: search_web → fetch_url (use URL from search result to get full content) and search_web → view_website (use URL to see rendered layout).

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

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