Skip to main content
AgentsUse

MCP server, by Firecrawl

Firecrawl MCP Server

Official Firecrawl MCP server that adds web scraping, search, crawling, and page interaction to Cursor, Claude, and other LLM clients for clean, agent-ready context.

stdio (local process)Streamable HTTP (remote)SSE (remote, legacy)Checked 2026-10-07Config checked

Quick facts

Transport
stdio (local process); Streamable HTTP (remote); SSE (remote, legacy)
Runs as
npx (firecrawl-mcp)
Review state
Config checked against the official source. Facts checked 2026-10-07. Not locally tested by AgentsUse.

What it does

A Model Context Protocol server that brings Firecrawl to MCP-compatible AI agents: search, scrape, and interact with the live web for clean, agent-ready context, per the README. Features listed include searching the web with full page content, scraping any URL into clean structured data, interacting with pages, deep research with an autonomous agent, credit usage checks, automatic retries and rate limiting, cloud and self-hosted support, and SSE support.

The README documents a hosted keyless free tier at https://mcp.firecrawl.dev/v2/mcp where scrape, search, and parse work without an API key (rate-limited), a search-only endpoint at https://mcp.firecrawl.dev/v2/mcp-search, running with npx, Streamable HTTP local mode, and Docker via ghcr.io/firecrawl/firecrawl-mcp-server. The full profile registers 27 tools with default settings.

Configuration

Hosted: connect to https://mcp.firecrawl.dev/v2/mcp with no setup for the keyless free tier, or with OAuth or Authorization Bearer FIRECRAWL_API_KEY for the full tool set. Local: npx -y firecrawl-mcp with FIRECRAWL_API_KEY env, Streamable HTTP local mode with HTTP_STREAMABLE_SERVER=true at http://localhost:3000/mcp, or Docker image ghcr.io/firecrawl/firecrawl-mcp-server.

MCP config (placeholders)
{"mcpServers":{"firecrawl-mcp":{"command":"npx","args":["-y","firecrawl-mcp"],"env":{"FIRECRAWL_API_KEY":"YOUR-API-KEY"}}}}

Source: https://github.com/firecrawl/firecrawl-mcp-server. Placeholders only. Replace them in your own environment, and never commit real keys to a repository.

Environment variableRequiredPurpose
FIRECRAWL_API_KEYYesRequired for Cloud API per README Configuration. Not required for the hosted keyless free tier scrape, search, and parse tools, which are rate-limited.
FIRECRAWL_API_URLOptionalSet to use a self-hosted Firecrawl API, per README Docker section.
HTTP_STREAMABLE_SERVEROptionalSet to true to run the server using Streamable HTTP locally instead of the default stdio transport, per README.

Verified tools and features

  • firecrawl_scrape Config checked

    Scrape a known URL and get its content as markdown or as JSON matching a schema you supply, per README tool guidance.

  • firecrawl_map Config checked

    Discover URLs on a site without fetching their content, per README tool guidance.

  • firecrawl_search Config checked

    Start from a query and get ranked web results, with optional scrapeOptions to fetch page content in the same call, per README tool guidance.

  • firecrawl_crawl Config checked

    Get content from many pages under a site, bounded by limit, includePaths/excludePaths, or maxDiscoveryDepth, per README tool guidance.

  • firecrawl_interact Config checked

    Click, type, or navigate on a page before reading it, using a url for a fresh page or a scrapeId to continue, per README tool guidance.

Security notes

Security notes
  • README hosted guidance: never put an API key in the server URL and never put an API key in an agent chat. Configure it directly in the client or secret manager, using the Authorization Bearer header for API-key connections.
  • Docker Streamable HTTP example publishes the port on 127.0.0.1 only because the server uses the container FIRECRAWL_API_KEY for requests without their own credential. Do not expose it more widely unless clients must supply their own key.
  • Other tools such as crawl, map, and agent still need an API key even on the hosted endpoint.

An MCP server runs with the permissions of the process you start it from. Scope credentials to the narrowest access the job needs, and review what the client will send before connecting anything sensitive.

Sources and freshness

Last checked 2026-10-07. Verification label: config checked, which means the configuration and feature list trace to the official source above. It does not mean AgentsUse ran the server. Spotted an error? Send a correction. Back to the MCP server index.