Firecrawl MCP Server: Tools, Setup and Security

The Firecrawl MCP server lets an AI app scrape pages into clean markdown or JSON and crawl or map whole sites. It can also search the web and click through pages in a real browser. Firecrawl publishes it on GitHub and hosts it at https://mcp.firecrawl.dev/v2/mcp. It also runs locally or self-hosted.

Checked against: Firecrawl MCP Server README on 28 October 2026. eSEOspace did not install or test this server.

At a glance

What it can do

The README documents these tool groups:

• Scrape: firecrawl_scrape turns a known URL into markdown or JSON matching a schema you supply.
• Map and crawl: firecrawl_map discovers a site's URLs. firecrawl_crawl gathers content from many pages, and firecrawl_check_crawl_status checks on a job.
• Search: firecrawl_search returns ranked web results. firecrawl_developer_search and firecrawl_gov_search search developer and US government sources.
• Research papers: four firecrawl_research_* tools search, inspect and read papers from sources such as PubMed and arXiv.
• Parse: firecrawl_parse reads PDFs and documents.
• Agent: firecrawl_agent runs multi-source research and returns structured data. You poll firecrawl_agent_status for results.
• Interact: firecrawl_interact clicks, types and navigates in a real browser. firecrawl_interact_stop ends the session.
• Monitor: firecrawl_monitor_* tools watch pages on a schedule and alert on meaningful changes.
• Credits: firecrawl_credit_usage shows credits left and past use.
• Feedback: firecrawl_search_feedback and firecrawl_feedback send feedback to Firecrawl. The README says you can opt out with environment variables.

How to connect it

For the hosted server with OAuth, point your client at:

https://mcp.firecrawl.dev/v2/mcp-oauth

For an API-key connection, keep the URL as https://mcp.firecrawl.dev/v2/mcp and set this header in your client's secret settings:

Authorization: Bearer <FIRECRAWL_API_KEY>

To run locally, the README gives:

env FIRECRAWL_API_KEY=fc-YOUR_API_KEY npx -y firecrawl-mcp

Its Claude Desktop config is:

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

The README also covers Cursor, Windsurf (now Devin Desktop), VS Code, Docker and a self-hosted Firecrawl instance via FIRECRAWL_API_URL. Firecrawl's hosted MCP setup guide has client-specific steps.

Security notes

Keys. The README is direct: "Never put an API key in the server URL. Never put an API key in an agent chat." Configure keys in the client or a secret manager. See MCP secret leakage.

Browser actions. firecrawl_interact can click, type and fill forms on live pages. Keep approval prompts on for it, and do not hand it logged-in sessions you would not give a stranger.

Smaller surface. The search-only endpoint at https://mcp.firecrawl.dev/v2/mcp-search exposes nine tools, and its search "fetches no page content." Use it when you only need search.

Stricter cloud mode. For self-hosted HTTP, CLOUD_SERVICE=true requires a credential on every request and makes scrape read-only. It binds to 0.0.0.0 by default, so put it behind your own network controls.

Untrusted pages. Scraped content can carry hidden instructions. Read prompt injection through MCP tool results and the MCP security center.

Pricing

Firecrawl's pricing page lists, as of October 2026:

The README notes that search with scrapeOptions bills each fetched result as a scrape.

Alternatives

• Bright Data MCP server: scraping through an unblocking proxy network, plus browser tools.
• Apify MCP server: thousands of ready-made scrapers as tools.
• Fetch MCP server: the free MCP reference server for simple page reads.

More on our MCP server profiles hub.

Next step

Firecrawl's map and scrape tools show what an AI agent can actually read on your site. If key pages come back thin, our MCP for SEO guide explains what to fix.

Frequently asked questions