Firecrawl MCP Server
Crawl, scrape, map, and search any website into clean markdown for your AI agent.
Updated
What is Firecrawl MCP?
The Firecrawl MCP server is a Model Context Protocol server that gives AI agents the ability to scrape, crawl, map, and search the web, returning clean markdown ready for an LLM to read.
Firecrawl is a web scraping API built for AI: it renders JavaScript, follows links, discovers a site's URL structure, and converts pages into structured markdown or JSON. The MCP server wraps that API so assistants like Claude and Cursor can pull live web data on demand — fetching current documentation, extracting structured fields, or running broader searches.
It is officially hosted by Firecrawl at the remote endpoint https://mcp.firecrawl.dev/v2/mcp, which works on the free tier without a key; adding your Firecrawl API key (via the URL path or the FIRECRAWL_API_KEY environment variable when running locally with npx) unlocks higher limits and the full tool set. The server can also be self-hosted against your own Firecrawl deployment.
Tools the Firecrawl MCP server exposes
| Tool | What it does |
|---|---|
| firecrawl_scrape | Extract the content of a single URL as markdown, JSON, or other formats. |
| firecrawl_batch_scrape | Scrape multiple known URLs in parallel in one batch operation. |
| firecrawl_map | Discover all indexed URLs on a website to map its structure. |
| firecrawl_crawl | Asynchronously crawl multiple pages of a site with depth and page limits. |
| firecrawl_check_crawl_status | Check the progress and results of an active crawl job. |
| firecrawl_search | Search the web and optionally scrape content from the results. |
| firecrawl_extract | Extract structured data from pages using LLM-powered analysis. |
| firecrawl_agent | Launch an autonomous research agent for complex, multi-source queries. |
Connect the Firecrawl MCP server
Claude Code
- 1
Add the remote server
Run: claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/v2/mcp — this connects to the hosted Firecrawl MCP server (works on the free tier with no key).
- 2
Add your API key (optional, for higher limits)
For full tool access, use the keyed URL instead: claude mcp add --transport http firecrawl https://mcp.firecrawl.dev/<FIRECRAWL_API_KEY>/v2/mcp. Get a key at firecrawl.dev/app/api-keys.
- 3
Verify
Run /mcp in Claude Code to confirm Firecrawl is connected, then ask the agent to scrape or search a URL.
Cursor
- 1
Open MCP settings
In Cursor, go to Settings → MCP and add a new server (Cursor 0.45.6+).
- 2
Configure the server
Add a server with command npx and args -y firecrawl-mcp, and set the environment variable FIRECRAWL_API_KEY to your Firecrawl API key. Alternatively point a remote/HTTP server at https://mcp.firecrawl.dev/v2/mcp.
- 3
Save and use
Save the config; Firecrawl's tools (scrape, crawl, map, search) become available to the Cursor agent.
Sample use cases
An agent needs the current contents of a documentation page or article
Firecrawl scrapes the URL and returns clean markdown the LLM can read directly.
A workflow must enrich a record with up-to-date web information
The agent runs a Firecrawl search and scrapes the top results to gather context.
You want a full site's pages indexed for a knowledge base
Firecrawl maps and crawls the site, returning structured markdown for every page.
Security & permissions
The hosted remote endpoint https://mcp.firecrawl.dev/v2/mcp requires no authentication on the free tier. To raise limits and access the full tool set, supply a Firecrawl API key — either embedded in the server URL path (https://mcp.firecrawl.dev/{FIRECRAWL_API_KEY}/v2/mcp) or as the FIRECRAWL_API_KEY environment variable when running locally via npx. Treat the API key as a secret. Once connected, the agent can scrape, crawl, and search arbitrary public web pages on your behalf and consume your account's credits, so scope which sites and workflows are allowed accordingly.
Using Firecrawl MCP with Gamut
In an automated Gamut workflow, an agent can call Firecrawl as a tool whenever a run is triggered by a schedule or an event — for example, crawling a competitor's site on a daily cron, scraping a list of URLs when a new record lands, or searching the web to enrich an incoming lead. Gamut manages the MCP connection so the agent simply invokes scrape, crawl, map, or search actions and receives clean markdown back to reason over, with no manual browser steps.
Frequently asked questions
What is the Firecrawl MCP server?
It is an official Model Context Protocol server from Firecrawl that lets AI agents scrape, crawl, map, and search the web, returning clean markdown or structured JSON. It wraps the Firecrawl web scraping API so clients like Claude and Cursor can pull live web data.
Is the Firecrawl MCP server free?
Yes — the hosted remote endpoint works on Firecrawl's free tier, which includes 1,000 credits per month with no card required. Paid plans (starting at Hobby $16/mo billed yearly) raise page limits and concurrency for heavier use.
How do I connect Firecrawl MCP to Claude or Cursor?
Point your client at the hosted URL https://mcp.firecrawl.dev/v2/mcp, or run it locally with npx -y firecrawl-mcp and a FIRECRAWL_API_KEY. In Claude Code use claude mcp add --transport http; in Cursor add it under Settings → MCP.
Is Firecrawl MCP hosted or self-hosted?
Both. Firecrawl officially hosts the remote server at mcp.firecrawl.dev, and you can also self-host the MCP server against your own Firecrawl deployment by configuring a custom Firecrawl API URL.