The server provides 5 tools with mostly complete JSON schemas and descriptions. However, descriptions are terse (averaging ~60 chars, well below the 194-char baseline), parameter descriptions are minimal or missing, and output schemas are undocumented. All tool names follow verb_noun convention (scrape_url, crawl_site, get_crawl_status, search_web, map_site), which is positive. Tool definitions are explicitly registered with schemas in internal/plugin.go, so they are not inferred. The HTTP client is properly abstracted (client/client.go), secret handling is correct (FIRECRAWL_API_KEY via env), and composition is reasonable (stateless tools, no state management). Key gap: output documentation, callers cannot predict response structure, and chaining opportunities are not clear.

• Expand each tool description to 100 - 250 chars. Include: WHAT does it do, WHEN to use it, expected latency, and any prerequisites. E.g., scrape_url: 'Fetch and parse a single URL into structured format (markdown, HTML, or links). Use for quick content extraction. Returns full page content by default; set only_main_content=true to skip headers/navs/footers. Useful before feeding content to analysis tools. Timeout: ~30s.'
• Document the output schema for every tool. Use a 'Returns' section: e.g., 'Returns: { content: string (markdown or HTML), metadata: { title, description, links: [{url, text}] }, status: enum(success|partial|error) }'. Include field descriptions.
• Replace example-based parameter descriptions with formal enum and range constraints. Change 'formats (e.g., markdown, html, rawHtml, links, screenshot)' to 'formats: array of enum ["markdown", "html", "rawHtml", "links", "screenshot"], default ["markdown"]'. Move enums into the JSON Schema if not already present.
• Add bounds documentation to numeric parameters: e.g., 'limit: number, 1 - 100 (default 10). Firecrawl rate limits prevent crawling >100 pages per call.'
• Include error handling guidance in each tool description: e.g., 'If the API returns 401 Unauthorized, verify FIRECRAWL_API_KEY is set. If 429 (rate limit), wait 60s and retry.'
• Document chaining requirements: e.g., 'get_crawl_status returns a list of URLs and status. Pass each URL to scrape_url to retrieve content' (if true). This guides LLM composition.
• Add a note to crawl_site and search_web about result limits: 'Results are capped at limit; if more exist, use pagination (if supported by Firecrawl) or make multiple calls with different queries.'