🔥 Firecrawl MCP Server

Operate a self-hosted Firecrawl instance from Claude via MCP.

LP: daisukehori.github.io/firecrawl-mcp

You can directly operate a self-hosted Firecrawl instance from Claude.ai / Claude Desktop / Claude Code via MCP (Model Context Protocol).

Equipped with 11 tools for web search, single/batch scraping, site crawling, sitemap retrieval, and LLM-structured extraction.

Architecture

Claude.ai ──MCP──▶ Vercel (firecrawl-mcp)
                        │
                        ▼
               Firecrawl (self-hosted)
               your-firecrawl.example.com

A Next.js app on Vercel handles the MCP protocol and relays requests to the self-hosted Firecrawl API. The server is stateless and does not store data or output logs.

Related MCP server: Firecrawl MCP Server

Tool List (11 Tools)

Scraping

Crawling

Search, Mapping, and Extraction

Quick Start (3 Steps)

Step 1: Prepare a self-hosted Firecrawl

Launch Firecrawl by following the Firecrawl Self-Host Guide.

git clone https://github.com/mendableai/firecrawl.git
cd firecrawl
docker compose up -d

It starts at http://localhost:3002 by default.

Step 2: Deploy the MCP Server

One-Click Deployment (Recommended)

Manual Deployment

git clone https://github.com/DaisukeHori/firecrawl-mcp.git
cd firecrawl-mcp
cp .env.example .env.local
# .env.local を編集
npm install
npm run dev

Step 3: Connect from Claude

Claude.ai (Web): Settings → MCP → Add:

• URL: https://your-firecrawl-mcp.vercel.app/api/mcp

• Header Name: x-api-key Value: Your MCP_API_KEY

Claude Desktop / Cursor / VS Code / Windsurf:

{
  "mcpServers": {
    "firecrawl": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://your-firecrawl-mcp.vercel.app/api/mcp"],
      "env": {
        "HEADER_x-api-key": "あなたのMCP_API_KEY"
      }
    }
  }
}

Claude Code:

claude mcp add --transport http firecrawl https://your-firecrawl-mcp.vercel.app/api/mcp \
  --header "x-api-key: あなたのMCP_API_KEY"

Environment Variables

Usage Examples

ユーザー: 「docs.firecrawl.dev の全ページをMarkdownで取得して」

AI: まずサイトマップを確認します。
    [firecrawl_map] → 47ページ検出
    
    クロールジョブを開始します。
    [firecrawl_crawl] → ジョブID: abc-123
    
    ... ステータス確認中 ...
    
    ✅ 47ページ全て取得完了しました。

ユーザー: 「この3つのURLから商品名と価格を抽出して」

AI: [firecrawl_extract]
    ✅ 構造化データ:
    {"name": "Product A", "price": 2980}
    {"name": "Product B", "price": 4980}
    {"name": "Product C", "price": 7980}

🔒 Security

• All communication is protected by HTTPS (TLS encryption)

• The server is stateless. Firecrawl URL and API keys are stored in Vercel environment variables and used only during requests

• No database on the server. It acts solely as a proxy to the Firecrawl API

• Source code is fully open

• Unauthorized access is prevented via MCP_API_KEY

FAQ

Q: Can I use it with the Firecrawl cloud version? → Yes. Set FIRECRAWL_API_URL=https://api.firecrawl.dev and enter your cloud API key in FIRECRAWL_API_KEY.

Q: The extract tool is not working → The self-hosted version requires LLM settings such as OPENAI_API_KEY on the Firecrawl side. Please check the Self-Host Guide.

Q: Batch scrape/crawl status remains as scraping → There is a few-second lag in status transitions in the self-hosted version. It will change to completed after polling 2-3 times. This is known behavior.

Q: Which version of Firecrawl is required? → A version that supports the v1 API (/v1/ endpoint). Using the latest version with Docker Compose will work fine.

Q: How is this different from the HubSpot MA MCP? → The HubSpot MA MCP is for HubSpot CRM/MA operations (128 tools). The Firecrawl MCP is for web scraping/crawling (11 tools). They serve different purposes and can be used together.

Tech Stack

Next.js 15 / TypeScript / MCP SDK / Vercel / Zod

License

MIT License

Related MCP Servers

A collection of MCP servers published by Hori. All are available for use from MCP clients such as Claude.ai / Cursor / ChatGPT, etc.