Firecrawl MCP

Firecrawl web scraping server — AI-powered web crawling, content extraction, and structured data collection at scale.

TRUSTED · Security verified

// Add to your client

{
  "mcpServers": {
    "firecrawl": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "<your-firecrawl-api-key>"
      }
    }
  }
}

{
  "mcpServers": {
    "firecrawl": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "<your-firecrawl-api-key>"
      }
    }
  }
}

claude mcp add firecrawl -e FIRECRAWL_API_KEY=<your-key> -- npx -y firecrawl-mcp

Paste into your client's MCP configuration file.

// Try it now

"Crawl https://example.com and return clean markdown of the 5 most important pages"

! Firecrawl is usage-based — crawling a 1000-page site adds up fast
! robots.txt and rate limits are respected — some sites return partial results
! Crawl depth and include/exclude patterns matter — a broad crawl can be expensive

Setup

Easy Setup

API key or simple config

Token usage

Light usage

// When to use

You want to ingest a whole documentation site as AI context
You need clean markdown from sites that are messy HTML
You're building RAG and need bulk-cleaned content
You want structured extraction (pricing, products, team members) from a site

// When NOT to use

You need one page fetch — use fetch-mcp (simpler and free)
You're scraping at massive scale — use dedicated infrastructure
You need interactive scraping (clicks, forms) — use Puppeteer or Browserbase

// Usage Scenarios

Documentation Ingestion

Crawl a whole docs site and get clean markdown for each page.

Example prompt

"Crawl https://docs.example.com and return clean markdown for every page under /guides"

Structured Extraction

Pull specific structured data from pages with AI-assisted selectors.

Example prompt

"Extract product name, price, and description from all product pages on example-shop.com"

Content Refresh

Re-crawl a site to get updated content for RAG indexes.

Example prompt

"Re-crawl our help center and return only pages that changed since last week"

// About

Plain English

Most websites weren't built for AI to read. They're full of pop-ups, navigation menus, ads, cookie banners, and code that only runs after the page is open in a browser. Hand a raw URL to a basic fetch tool and you often get back a mess. Firecrawl is built for the opposite problem: feed it a website and it returns clean, AI-ready text. It runs a real browser behind the scenes, waits for everything to load, strips out the noise, and gives you the actual content as Markdown — the same format AI assistants prefer. It can grab a single page, walk an entire documentation site, or pull structured data (like a list of products with prices) according to a shape you describe. For a builder, this turns 'I wish my AI knew everything on this docs site' into a one-line request. For research, it turns a competitor's site or a news section into a clean source you can summarize. The trade-off versus a basic fetch tool: it costs money per page on the hosted plan, but it handles the modern web that simpler tools choke on.

The Firecrawl MCP server enables advanced web scraping and crawling through AI assistants. Crawl entire websites, extract structured content, convert pages to clean Markdown, and collect data at scale with built-in JavaScript rendering and anti-bot bypass.

Maintained by Firecrawl with a production API. The preferred choice for large-scale, reliable web data extraction.

GitHub 5.2kUpdated Feb 14, 2025MIT

View on GitHub Official Site