Single tool 'scrape' has a valid Zod schema with URL validation, but critical gaps undermine quality. Tool description is missing entirely (0 chars visible in registration). Parameter description 'Please provide a valid URL' is generic and under 20 chars, violating baseline standards. No output schema documented, LLMs cannot predict return structure. Error handling returns unstructured text ('Error: ...') without recovery guidance. No pagination, rate limiting, or result size caps despite calling external API. Hardcoded API credentials in source code (Bearer token visible) is a critical security violation. Tool does one thing (scrape) but lacks production-grade documentation and safety patterns.

• Add a tool description to server.tool() call: 'Scrape and extract markdown content from a URL. Removes headers, footers, ads, and media. Returns sanitized markdown text. Use this to teach the LLM about web content.'
• Move API credentials to environment variables (process.env.FIRECRAWL_API_KEY, process.env.FIRECRAWL_API_URL) and inject at runtime, not in source.
• Document output schema: { success: boolean, markdown: string, error?: string }. Include this in tool description or as a separate schema annotation.
• Enhance error handling: return { isError: true, errorType: 'retryable'|'user_fixable'|'fatal', message: '...', suggestion: '...' } so LLMs know whether to retry or ask user.
• Add timeout (e.g., 30s) and result size cap (e.g., max 50KB markdown) to prevent runaway calls and context explosion.