Single tool (fetch_url_content) with basic definition quality. Tool name follows verb_noun convention (fetch_url). Description is present (35 chars) but minimal and lacks context about when to use it, prerequisites, and return format. Input schema is present with proper JSON Schema and Zod validation (url string with URL format constraint), but output schema is completely undocumented, the description does not specify what format/structure is returned. No parameter descriptions beyond the name. Error handling is minimal, fetch failures throw generic errors without recovery guidance. No security considerations documented (API key injection via environment variable is good practice, but not mentioned in tool description). Tool does exactly one thing (good composition), but lacks LLM-ready description optimizations.

• Expand tool description to 80-150 chars, including: 'Fetches the full text content of a URL and converts it to markdown using the Jina Reader API. Returns plain markdown string. Use this to extract readable content from web pages, articles, and documents. Requires optional JINA_API_KEY environment variable for higher rate limits.'
• Document the output schema explicitly: 'Returns a string containing the markdown-formatted content of the URL. Empty string if the page has no readable text.'
• Add description to the url parameter: 'The URL to fetch content from. Must be a valid HTTP or HTTPS URL. Maximum 2000 characters. Examples: https://example.com, https://en.wikipedia.org/wiki/MCP'
• Implement structured error handling with recovery guidance: if response.status === 404, throw 'URL not found (404). Check that the URL is correct and accessible.' If response.status >= 500, throw 'Service error (HTTP 500+). Retry after a short delay.' If fetch times out, throw 'Request timed out after 30 seconds. The server may be slow or offline. Try again later.'
• Add a timeout parameter or constant (e.g., 30s) to the fetch call to prevent indefinite hangs.
• Document what happens when content cannot be extracted (empty response, binary file, etc.), does the tool return empty string, throw error, or return metadata?