Firecrawl MCP 🔌 MCP server Freemium

What it does

Firecrawl MCP connects an agent to Firecrawl's web data API. It turns messy web pages into clean Markdown or structured JSON that a model can use directly. A coding agent can use it to scrape a page before copying its layout or content, crawl a site's docs, search the web with full page content in the results, or pull structured data against a schema you give it. It can also interact with pages before reading them. The server is MIT licensed and works with Firecrawl's cloud or a self-hosted Firecrawl instance.

Tools it gives your agent

• firecrawl_scrape, firecrawl_map, firecrawl_crawl with a status check, and firecrawl_search
• firecrawl_developer_search, over an index of GitHub issues, pull requests, READMEs and docs
• firecrawl_interact, to click, type and navigate before reading a page
• firecrawl_agent, for deep research, plus firecrawl_parse for files
• firecrawl_monitor_*, for scheduled checks with diffs
• Research tools for finding and reading papers

Works with

It is available as a hosted remote server at https://mcp.firecrawl.dev/v2/mcp, or locally through npx firecrawl-mcp over stdio or local Streamable HTTP. The README has setup steps for Cursor, Windsurf and VS Code, and Firecrawl's docs name Claude Code and Codex. Any MCP client works.

How to connect

Hosted, with an API key in a header:

{ "mcpServers": { "firecrawl": { "type": "http", "url": "https://mcp.firecrawl.dev/v2/mcp", "headers": { "Authorization": "Bearer <FIRECRAWL_API_KEY>" } } } }

Local:

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

Cost and limits

At the time of writing (September 2026), the Free plan gives 1,000 credits a month with no card needed, and a basic scrape of one page costs one credit. Hobby costs $19 a month for 5,000 credits, and larger plans go up to one million credits. Without an API key, the hosted server still allows rate-limited scrape, search and parse. Crawl, map and agent always need a key.

Safety notes

Never put your API key in the server URL or paste it into chat. Store it in your client's secret settings or header config. If you only need search, the hosted endpoint https://mcp.firecrawl.dev/v2/mcp-search exposes six read-only tools and never fetches page content. Scraped pages are untrusted input and can contain prompt injection, so review what the agent does with them.