# Firecrawl Web

Use these tools when web access should go through Firecrawl only:

- Use `web_search` for discovery, current information, and open-ended queries where you do not already know the target URL. Keep `limit` small by default.
- Use `fetch_content` when the user provides a URL or after `web_search` finds a relevant URL. Prefer `formats: ["markdown"]` for clean agent-readable content.
- Use `extract_structured` when the user asks for structured data such as tables, lists, fields, prices, plans, contacts, or comparisons from one or more URLs. Provide a `schema` when the desired shape is known; otherwise provide a clear `prompt`.

Security and cost notes:

- Only HTTP/HTTPS URLs are allowed.
- Wildcards like `https://example.com/*` may consume many Firecrawl credits; warn the user before broad extraction.
- Never reveal or print the Firecrawl API key.