Firecrawl MCP Server: Web Scraping and Search

Overview

Firecrawl MCP Server is a Model Context Protocol (MCP) server that connects LLM clients (Cursor, Claude Desktop, and other MCP-capable tools) to Firecrawl’s web scraping, crawling, and extraction capabilities. It exposes Firecrawl as an MCP server so agents and assistants can request live web content, structured extraction, search, and batch crawls as part of a model’s context window.

For developers building LLM-based workflows that need reliable, LLM-friendly web context, this server provides browser-backed scraping, automatic retries and rate-limiting, SSE streaming, and cloud browser sessions — all available over the standard MCP transports. It can run locally via npx/npm or be deployed in environments where MCP servers are supported.

Features

• Web scraping and site crawling with content extraction
• Search and discovery across crawled data
• Batch scraping / deep research pipelines
• Cloud browser sessions and agent-browser automation (JavaScript rendering)
• Automatic retries, backoff and rate limiting
• SSE (Server-Sent Events) support for streaming results
• Run locally (stdio) or as HTTP streamable MCP server
• Cloud-hosted and self-hosted modes
• Integrates with MCP clients such as Cursor, Claude Desktop, and other LLM tools

Installation / Configuration

Prerequisites: Node.js and npm.

Quick run with npx (recommended for local / dev):

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

Global install:

npm install -g firecrawl-mcp
firecrawl-mcp
# or:
env FIRECRAWL_API_KEY=fc-YOUR_API_KEY firecrawl-mcp

Run in Streamable HTTP local mode (exposes an HTTP MCP endpoint):

env HTTP_STREAMABLE_SERVER=true FIRECRAWL_API_KEY=fc-YOUR_API_KEY npx -y firecrawl-mcp
# Default local URL:
# http://localhost:3000/mcp

Windows note: if setting env vars inline causes issues, use a cmd wrapper:

cmd /c "set FIRECRAWL_API_KEY=your-api-key && npx -y firecrawl-mcp"

Environment variables (summary):

Cursor configuration (example JSON for Cursor v0.48.6+):

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

VS Code (User Settings JSON snippet):

{
  "mcp": {
    "inputs": [
      {"type": "promptString","id": "apiKey","description": "Firecrawl API Key","password": true}
    ],
    "servers": {
      "firecrawl": {
        "command": "npx",
        "args": ["-y", "firecrawl-mcp"],
        "env": {"FIRECRAWL_API_KEY": "${input:apiKey}"}
      }
    }
  }
}

Legacy / automated install (Smithery):

npx -y @smithery/cli install @mendableai/mcp-server-firecrawl --client claude

Available Resources

• GitHub repo: https://github.com/firecrawl/firecrawl-mcp-server
• Firecrawl API keys & dashboard: https://www.firecrawl.dev/app/api-keys
• Local Streamable HTTP endpoint (when enabled): http://localhost:3000/mcp
• MCP transports supported: stdio (default), Streamable HTTP (SSE), and MCP-compatible client integrations

Use Cases

• Research summary with live web context:
  • Ask an LLM agent to “summarize the latest posts on example.com and extract main claims.” The agent uses Firecrawl MCP to fetch and extract article texts, then produces a contextualized summary.
• Batch extraction pipeline:
  • Provide a list of 1,000 URLs to scrape; Firecrawl MCP runs a batch crawl, performs JS rendering where needed, extracts structured fields (title, author, publish date, body), and returns results in chunks or via SSE.
• Integrating scraping into conversational flows:
  • A Claude agent can call the MCP server to fetch a web page during a conversation, then quote or paraphrase the text with citations.
• Competitive research and site discovery:
  • Run a discovery crawl across a domain, index found pages, and perform search queries against the discovered content from the LLM for fast referencing.
• Headless automation and agent-browser tasks:
  • Use cloud browser sessions to interact with pages that require dynamic interactions (forms, navigation), letting automated agents perform multi-step collection workflows.

Quick tips & troubleshooting

• Ensure FIRECRAWL_API_KEY is set for cloud usage; local self-hosted deployments may require different configuration.
• If you need streaming results in MCP clients that support SSE, enable HTTP_STREAMABLE_SERVER and point the client to http://localhost:3000/mcp.
• For Windows, use the cmd wrapper pattern shown above when launching with inline env vars.
• If an MCP client can’t see your server, verify the transport (stdio vs HTTP) matches the client’s MCP configuration.

Next steps

• Clone the repository to review source and contribute: https://github.com/firecrawl/firecrawl-mcp-server
• Get an API key and test the server locally with npx (quickest path)
• Add the server to your MCP client (Cursor, VS Code, Claude Desktop) and experiment with real-time web context in your LLM applications

This server is focused on turning web content into reliable, LLM-friendly context — ideal for content extraction, search augmentation, and research workflows that require live web data.