This MCP server has severe definition quality issues. Only 2 tools are exposed, both with minimal descriptions (under 100 chars). Input schemas are present but lack parameter descriptions, critical for LLM reasoning. The 'search_companies' tool description is generic and doesn't explain when to use it vs alternatives. 'scrape_company_page' lacks any guidance on expected output format or error cases. No output schemas are documented. Parameter descriptions are missing entirely (e.g., 'query' and 'num_results' have descriptions, but they are minimal). The server appears to be a wrapper around Firecrawl API but exposes no error handling guidance, no pagination support, and no structured output documentation. Tool naming is acceptable (verb_noun pattern), but the implementation quality is far below production baseline.

• Add detailed output schemas for both tools. Document what fields search_companies returns (e.g., list of {url, title, snippet, metadata}) and what scrape_company_page returns (e.g., {markdown, title, url, status_code}).
• Expand tool descriptions to 150-250 characters. Explain: WHAT does it do? WHEN should the LLM call it? WHAT does it return? Example: 'Search for companies and developer tools by name or category. Returns up to num_results results with URLs and snippets. Use this to discover tools; then call scrape_company_page to extract full details.'
• Add parameter descriptions for all inputs. For 'query': 'Search query string (e.g., "API gateway", "monitoring tool"). Supports company names, tool categories, and keywords.' For 'num_results': 'Number of results to return (1-50, default 5). Higher values increase latency.'
• Document error cases and recovery paths. E.g., 'If no results found, try a broader query or different keywords. If scrape fails, the URL may be blocked or invalid, suggest alternative sources.'
• Add pagination support: accept 'offset' or 'cursor' parameter and return 'total_count' and 'next_cursor' in response so agents can iterate over large result sets without re-querying.