Firecrawl MCP Server

Official Firecrawl MCP server for web scraping, crawling, search, and LLM-powered structured data extraction. Turns websites into clean, LLM-ready data.

Data & Enrichment by Mendable AI (Firecrawl) API Key active

GitHub Docs npm: firecrawl-mcp

Overview

Firecrawl MCP Server is the official Model Context Protocol integration for Firecrawl, a hosted web scraping and crawling API maintained by Mendable AI. It exposes Firecrawl's capabilities as tools that any MCP client (Claude Desktop, Cursor, VS Code, etc.) can call, letting an agent fetch single pages, run large-scale crawls, search the web, and extract structured JSON from arbitrary sites without writing scraper code.

The server wraps the Firecrawl API with tools for single-URL scraping, batch scraping, sitemap-style URL discovery (firecrawl_map), recursive crawling with depth/limit controls, web search with optional full-page content, schema-based structured extraction powered by an LLM, and an autonomous research agent that performs multi-step browsing tasks asynchronously. It supports both the Firecrawl cloud service and self-hosted Firecrawl instances via a configurable API URL.

Notable features include automatic retry with exponential backoff for rate limiting, credit usage monitoring with configurable warning thresholds, JSON schema-based extraction, and streamable HTTP transport in addition to stdio. The package is published as firecrawl-mcp on npm and can be run instantly via npx.

Tools

Tool	Description
`firecrawl_scrape`	Extract content from a single URL, returning markdown, HTML, or structured JSON.
`firecrawl_batch_scrape`	Scrape multiple known URLs efficiently in a single asynchronous job.
`firecrawl_check_batch_status`	Check progress and retrieve results of a batch scrape job.
`firecrawl_map`	Discover indexed URLs on a website (sitemap-style URL discovery).
`firecrawl_search`	Run a web search and optionally extract content from result pages.
`firecrawl_crawl`	Recursively crawl a site with depth and page limit controls (asynchronous).
`firecrawl_check_crawl_status`	Monitor the progress of a crawl job and retrieve results.
`firecrawl_extract`	Extract structured data from one or more URLs using an LLM and a JSON schema or prompt.
`firecrawl_agent`	Launch an autonomous research agent that performs multi-step browsing (async).
`firecrawl_agent_status`	Check the status and retrieve results of a running agent job.
`firecrawl_interact`	Interactive page automation including click, type, and navigate actions.

Setup Guide

Prerequisites

A Firecrawl API key from firecrawl.dev/app/api-keys (free tier available, paid plans for higher credit limits)
Node.js installed locally

Install

Run directly with npx (no install needed):

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

Or install globally:

npm install -g firecrawl-mcp

Claude Desktop / Cursor config

Add to your MCP client config (e.g. claude_desktop_config.json):

{
  "mcpServers": {
    "mcp-server-firecrawl": {
      "command": "npx",
      "args": ["-y", "firecrawl-mcp"],
      "env": {
        "FIRECRAWL_API_KEY": "YOUR_API_KEY_HERE",
        "FIRECRAWL_RETRY_MAX_ATTEMPTS": "5",
        "FIRECRAWL_RETRY_INITIAL_DELAY": "2000",
        "FIRECRAWL_RETRY_MAX_DELAY": "30000"
      }
    }
  }
}

Environment variables

FIRECRAWL_API_KEY (required): Firecrawl cloud API key
FIRECRAWL_API_URL (optional): Override endpoint for self-hosted Firecrawl
FIRECRAWL_RETRY_MAX_ATTEMPTS (default 3)
FIRECRAWL_RETRY_INITIAL_DELAY ms (default 1000)
FIRECRAWL_RETRY_MAX_DELAY ms (default 10000)
FIRECRAWL_RETRY_BACKOFF_FACTOR (default 2)
FIRECRAWL_CREDIT_WARNING_THRESHOLD (default 1000)
FIRECRAWL_CREDIT_CRITICAL_THRESHOLD (default 100)

Streamable HTTP mode

To run over streamable HTTP instead of stdio:

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

Use Cases

Scrape competitor pricing pages or product catalogs into structured JSON for analysis without writing custom parsers.
Crawl entire documentation sites or knowledge bases and convert them to clean markdown for ingestion into a RAG pipeline.
Run web searches inside an agent workflow and feed the full-text results back into the LLM for synthesis.
Extract structured data (company info, contact details, specs) from a list of URLs using a JSON schema and firecrawl_extract.
Kick off long-running research agents that browse multiple sources and return synthesized findings via firecrawl_agent.

Example Prompts

"Scrape the pricing page at example.com/pricing and return tiers, monthly price, and included features as JSON."
"Crawl docs.stripe.com up to depth 2, limit 200 pages, and save the markdown for each page."
"Search the web for 'best open-source vector databases 2026' and extract the top 5 articles' main content."
"Use firecrawl_extract on these 20 startup URLs and pull out company name, HQ city, and funding stage."
"Launch a Firecrawl agent to research recent EU AI Act enforcement actions and return a structured summary."

Pros

Official server maintained by Firecrawl/Mendable AI, kept in sync with the underlying API.
Broad capability set: single-page scrape, batch, crawl, map, search, LLM extraction, and autonomous agent in one server.
Built-in retry with exponential backoff and credit usage alerts to prevent surprise bill spikes.
Supports both the hosted Firecrawl cloud and self-hosted instances via FIRECRAWL_API_URL.

Limitations

Requires a Firecrawl API key and consumes credits; heavy crawling on paid plans can get expensive.
Crawl, batch, and agent operations are asynchronous, requiring follow-up check_status calls and adding orchestration complexity.
Some legacy firecrawl_browser_* session tools are deprecated, which can be confusing if following older examples.

Alternatives

Apify MCP Server: access thousands of pre-built scrapers (Actors) for specific sites.
Browserbase MCP: cloud headless browsers for interactive scraping and automation.
Playwright MCP: Microsoft's official browser automation server for fully scripted scraping.