Trafilatura MCP provides a single, well-defined tool with clear naming and structured schema. The tool name 'fetch_and_extract' starts with verbs but combines two operations (FETCH + EXTRACT), which violates single-responsibility principle. The description is adequate (158 chars, within baseline 34 - 392 range). Input schema is comprehensive with all three parameters properly typed and described. However, the tool lacks documentation of its output schema structure, the response is a JSON string from Trafilatura, but no schema is provided explaining what fields the LLM should expect. Error handling is present but generic (McpError with INTERNAL_ERROR code). The tool is read-only, which is appropriate.

• Split 'fetch_and_extract' into two separate tools: 'fetch_webpage' (returns raw HTML with metadata like status, content-type, encoding) and 'extract_content' (takes HTML, returns structured content). This enables agents to retry or inspect raw content if extraction fails.
• Document the output schema explicitly. Trafilatura returns JSON with keys like 'main_content', 'meta', 'comments'. Provide a JSON Schema fragment or description of these fields so LLMs know what to expect: 'Returns {main_content: string|null, meta: {author, date, source, ...}, comments: string|null, ...}'
• Enhance error messages with recovery guidance. Example: 'Failed to fetch URL (status code 403). The URL may require authentication, be behind a paywall, or block automated requests. Try a different URL or check if the site allows web scraping.'
• Add a timeout parameter (default 30s) with clear constraint: 'Timeout in seconds (5 - 120, default 30). Increase for slow or heavily-loaded sites.'
• Add URL validation: reject localhost, 127.0.0.1, private IP ranges (10.0.0.0/8, 172.16.0.0/12, 192.168.0.0/16) to prevent SSRF attacks. Document this in the description: 'URL must be a public-facing website (not localhost or internal IP).'
• Return structured JSON with documented fields instead of passing raw trafilatura output. This normalizes the response and prevents token bloat from Trafilatura's verbose metadata.
• Document what each parameter actually controls in the description. For 'include_comments': 'If true, extract user comments/discussions from the page (may add significant content). Useful for blogs and forums; set false for news articles to reduce noise.'

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

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