Firecrawl MCP: scrape pages, crawl sites, extract structured data

A basic fetch tool gives up on JavaScript-heavy pages, docs sites that run to dozens of pages, or tables you want as structured fields. Firecrawl MCP is Firecrawl’s official server. It passes single-page scrapes, site crawls, link maps and schema extraction to Firecrawl’s backend, and the agent gets only the results. It had about 7.6k stars as of 2026-10-06.

What it does

• Single pages: firecrawl_scrape turns a known URL into Markdown, or into JSON that matches a schema you supply.
• Crawls and maps: firecrawl_map lists a site’s URLs without fetching them, firecrawl_crawl pulls many pages within limits such as limit and includePaths, and firecrawl_check_crawl_status tracks progress.
• Search plus scrape: firecrawl_search starts from a query and can fetch page content in the same call.
• Interaction and monitoring: firecrawl_interact clicks and types before reading, firecrawl_monitor_* checks pages on a schedule and reports diffs, and firecrawl_agent runs autonomous research.

Who it’s for

• Developers feeding a whole docs site, blog or product catalog to an agent for RAG or analysis.
• Data and growth teams extracting prices, job listings or other structured fields at scale.

Setup

Needs a Firecrawl API key (or FIRECRAWL_API_URL for a self-hosted instance). For Claude Desktop:

{
  "mcpServers": {
    "mcp-server-firecrawl": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "YOUR_API_KEY_HERE"
      }
    }
  }
}

You can also connect to the hosted endpoint https://mcp.firecrawl.dev/v2/mcp; its keyless tier allows scrape, search and parse with rate limits.

Our take

For reading a handful of pages, the official Fetch server is enough. Firecrawl earns its place with scale and structure: JavaScript rendering, whole-site crawls, schema extraction and scheduled monitoring, none of which a fetch tool does alone. The catch is that it mostly runs on paid cloud credits. crawl, map and agent need a key, large crawls burn credits fast, and firecrawl_credit_usage shows what’s left. Self-hosting avoids the bill but means running Firecrawl itself. Before crawling someone else’s site at scale, check its terms of service and robots rules. Licensed MIT.