Firecrawl MCP Server

Firecrawl is what you graduate to when simple fetching stops being enough. Its MCP server wraps a hosted extraction platform that renders JavaScript, rotates proxies, retries intelligently and returns clean, LLM-ready markdown — from a single page, a batch of URLs, or an entire site crawled link by link.

The tool spread tells the story. scrape handles one URL with full rendering, so SPAs and lazy-loaded content actually show up. map discovers all the URLs on a site fast — handy on its own for auditing site structure. crawl walks a whole site with depth and page limits and returns everything as markdown, which is how you turn a documentation site into a knowledge base in one command. search runs a web search and optionally scrapes the results in the same call, collapsing the find-then-read loop into one step. And extract pulls structured data — you describe a schema, it returns JSON matching it, from pages that were never meant to be machine-readable.

That last one is quietly the killer feature for real work: "get me name, price and rating for every product on these pages, as JSON" is a data-engineering task that becomes a sentence.

It's a hosted service with credits — the free tier is enough to evaluate honestly, and self-hosting the open-source engine is possible if you'd rather run your own. Against the alternatives: Fetch is free and fine for static pages; Playwright gives you interactive control; Apify has site-specific specialists. Firecrawl's lane is bulk, reliable, rendering-aware extraction with minimal ceremony, and in that lane it's currently the one to beat.

Cost and quota notes

• map is cheap; crawl is not. Map the domain first, look at the URL list, then crawl with a limit and path filters set — pointing crawl at a site root and hoping is how people burn a month of credits in an afternoon.
• Crawls are asynchronous: the call returns a job you poll for status. Big sites take minutes, so expect the assistant to check back rather than block on one call.
• onlyMainContent defaults to on and strips nav, sidebars and footers. Switch it off only when the thing you actually want lives in the page chrome — pricing footers, for instance.
• extract runs a model over each page, so it costs meaningfully more than scrape. If a CSS selector or a regex over the markdown would do, scrape and parse it yourself.