Firecrawl MCP server: give your AI live web data (2026)

A Firecrawl MCP server gives an AI client tools to scrape a page, crawl a site and search the live web — returning clean markdown an LLM can actually use. Firecrawl maintains the official server; most people are connected in under three minutes. Here's the setup.

Step 1 — get an API key

Create an account and grab a key from firecrawl.dev/app/api-keys. Keys are prefixed fc-.

Step 2 — configure the client

The official package is firecrawl-mcp:

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

Claude Code one-liner:

claude mcp add --transport stdio --env FIRECRAWL_API_KEY=fc-YOUR_API_KEY firecrawl -- npx -y firecrawl-mcp

On Windows, if env-style launching fails, use cmd /c "set FIRECRAWL_API_KEY=fc-... && npx -y firecrawl-mcp". Restart the client; see MCP config file location.

Step 3 — verify

Ask: "Scrape example.com and give me the main heading and first paragraph as markdown." Clean content back confirms it works.

Common problems

401 Unauthorized — the key is missing, wrong, or out of credits; check the dashboard.
Large crawl stalls — crawling a whole site burns credits and time; start with single-page scrape before crawl.
No tools in client — Node 18+ for npx; see Cursor MCP not working.

Security

Firecrawl fetches arbitrary URLs on your behalf and bills against your account — scope which sites you point it at, watch the credit threshold, and keep the key out of shared configs. See MCP security best practices.

Firecrawl vs the alternatives

For research-grade search use Exa or Tavily; for large managed scraping pipelines compare Apify and Bright Data. See best MCP for browser automation.

Going further

See the Firecrawl agent profile, browse the search category, or build a content-writer loadout.