🕷️ Crawl4AI MCP Server

MCP (Model Context Protocol) server for Crawl4AI - Universal web crawling and data extraction for AI agents.

Integrate powerful web scraping capabilities into Claude, ChatGPT, and any MCP-compatible AI assistant.

📑 Table of Contents

• 🐳 Quick Start with Docker (Recommended)
• 📦 Alternative Installation Methods
• 🛠️ Available Tools
• 🚀 Usage
• ⚙️ Configuration
• 🤝 Contributing
• 📄 License

🐳 Quick Start with Docker (Recommended)

✨ Docker is the preferred way to run Crawl4AI MCP Server - everything is pre-installed and ready to go!

Option 1: Docker Hub Image (Latest)

# SSE mode (for web interfaces) - DEFAULT
docker run --rm -p 3001:9001 \
  -e CRAWL4AI_ENDPOINT="https://your-crawl4ai-server.com" \
  -e CRAWL4AI_BEARER_TOKEN="your-optional-token" \
  stgmt/crawl4ai-mcp:latest crawl4ai-mcp --sse

# HTTP mode (for REST API)
docker run --rm -p 3000:3000 \
  -e CRAWL4AI_ENDPOINT="https://your-crawl4ai-server.com" \
  -e CRAWL4AI_BEARER_TOKEN="your-optional-token" \
  stgmt/crawl4ai-mcp:latest crawl4ai-mcp --http --port 3000

# STDIO mode (for Claude Desktop)
docker run --rm -it \
  -e CRAWL4AI_ENDPOINT="https://your-crawl4ai-server.com" \
  -e CRAWL4AI_BEARER_TOKEN="your-optional-token" \
  stgmt/crawl4ai-mcp:latest crawl4ai-mcp --stdio

Option 2: Build from Source (Latest fixes)

# Clone and build
git clone https://github.com/stgmt/crawl4ai-mcp.git
cd crawl4ai-mcp
docker build -t crawl4ai-mcp:local .

# Run SSE mode
docker run --rm -p 3001:9001 \
  -e CRAWL4AI_ENDPOINT="https://your-crawl4ai-server.com" \
  -e CRAWL4AI_BEARER_TOKEN="your-optional-token" \
  crawl4ai-mcp:local crawl4ai-mcp --sse

With Claude Desktop (Docker)

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "crawl4ai": {
      "command": "docker",
      "args": [
        "run", "--rm", "-i",
        "-e", "CRAWL4AI_ENDPOINT=https://your-crawl4ai-server.com",
        "-e", "CRAWL4AI_BEARER_TOKEN=your-optional-token",
        "stgmt/crawl4ai-mcp:latest",
        "crawl4ai-mcp", "--stdio"
      ]
    }
  }
}

📦 Alternative Installation Methods

NPM Package

# Install globally
npm install -g crawl4ai-mcp-sse-stdio

# Run in different modes (set CRAWL4AI_ENDPOINT first)
export CRAWL4AI_ENDPOINT="https://your-crawl4ai-server.com"
npx crawl4ai-mcp --stdio
npx crawl4ai-mcp --sse --port 3001
npx crawl4ai-mcp --http --port 3000

Python Package (PyPI)

# Install from PyPI
pip install crawl4ai-mcp

<<<<<<< HEAD
# Set required endpoint and run
export CRAWL4AI_ENDPOINT="https://your-crawl4ai-server.com"
crawl4ai-mcp --stdio
=======
# Run with command line arguments (recommended)
crawl4ai-mcp --stdio --endpoint https://your-crawl4ai-server.com
crawl4ai-mcp --http --port 3000 --endpoint https://your-crawl4ai-server.com
crawl4ai-mcp --sse --port 3001 --endpoint https://your-crawl4ai-server.com

# With optional bearer token
crawl4ai-mcp --stdio --endpoint https://your-crawl4ai-server.com --bearer-token your-token

From NPM (Alternative)

npm install -g crawl4ai-mcp-sse-stdio

# Run with command line arguments (recommended)
npx crawl4ai-mcp --stdio --endpoint https://your-crawl4ai-server.com
npx crawl4ai-mcp --http --port 3000 --endpoint https://your-crawl4ai-server.com
npx crawl4ai-mcp --sse --port 3001 --endpoint https://your-crawl4ai-server.com

# With optional bearer token
npx crawl4ai-mcp --stdio --endpoint https://your-crawl4ai-server.com --bearer-token your-token

From Source

git clone https://github.com/stgmt/crawl4ai-mcp.git
cd crawl4ai-mcp
pip install -e .

With Claude Desktop (Non-Docker)

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "crawl4ai": {
      "command": "crawl4ai-mcp",
      "args": ["--stdio"],
      "env": {
        "CRAWL4AI_ENDPOINT": "https://your-crawl4ai-server.com",
        "CRAWL4AI_BEARER_TOKEN": "your-optional-token"
      }
    }
  }
}

🛠️ Available Tools

1. crawl - Full Web Crawling

Extract complete content from any webpage.

{
  "name": "crawl",
  "arguments": {
    "url": "https://example.com",
    "wait_for": "css:.content",
    "timeout": 30000
  }
}

2. md - Markdown Extraction

Get clean markdown content from webpages.

{
  "name": "md", 
  "arguments": {
    "url": "https://docs.example.com",
    "clean": true
  }
}

3. html - Raw HTML

Retrieve raw HTML content.

{
  "name": "html",
  "arguments": {
    "url": "https://example.com"
  }
}

4. screenshot - Visual Capture

Take screenshots of webpages.

{
  "name": "screenshot",
  "arguments": {
    "url": "https://example.com",
    "full_page": true
  }
}

5. pdf - PDF Generation

Convert webpages to PDF.

{
  "name": "pdf",
  "arguments": {
    "url": "https://example.com",
    "format": "A4"
  }
}

6. execute_js - JavaScript Execution

Execute JavaScript on webpages.

{
  "name": "execute_js",
  "arguments": {
    "url": "https://example.com",
    "script": "document.title"
  }
}

🚀 Usage

The crawl4ai-mcp server supports multiple transport modes and provides comprehensive web crawling capabilities through the Model Context Protocol.

Basic Commands

# HTTP mode (recommended for testing)
crawl4ai-mcp --http --port 3000 --endpoint https://your-crawl4ai-server.com

# SSE mode (Server-Sent Events)
crawl4ai-mcp --sse --port 3001 --endpoint https://your-crawl4ai-server.com

# STDIO mode (for MCP clients)
crawl4ai-mcp --stdio --endpoint https://your-crawl4ai-server.com

# With optional bearer token
crawl4ai-mcp --http --port 3000 --endpoint https://your-crawl4ai-server.com --bearer-token your-token

With Custom Endpoint

# Using custom Crawl4AI endpoint with bearer token
crawl4ai-mcp --http --port 3000 \
  --crawl4ai-endpoint "https://your-server.com" \
  --bearer-token "your-token"

See the Python Integration section for detailed code examples.

⚙️ Configuration

Environment Variables

# REQUIRED: Crawl4AI endpoint URL
export CRAWL4AI_ENDPOINT="https://your-crawl4ai-server.com"

# OPTIONAL: Bearer authentication token
export CRAWL4AI_BEARER_TOKEN="your-api-token"

Parameter Requirements:

• CRAWL4AI_ENDPOINT - Required - The URL of your Crawl4AI server instance
• CRAWL4AI_BEARER_TOKEN - Optional - Bearer token for authenticated API access

Command Line Options

crawl4ai-mcp --help

Options:
  --stdio              Run in STDIO mode for MCP clients
  --sse                Run in SSE mode for web interfaces (default)
  --http               Run in HTTP mode
  --endpoint ENDPOINT  Crawl4AI API endpoint URL (REQUIRED)
  --bearer-token TOKEN Bearer authentication token (OPTIONAL)
  --version, -v        Show version

🐍 Python Integration Example

Here's how to integrate the MCP server with your Python application using HTTP mode with bearer token authentication:

import asyncio
import aiohttp
import json

async def test_crawl4ai_mcp():
    """
    Example: Using Crawl4AI MCP server via HTTP with bearer token
    """
    # Server configuration
    server_url = "http://localhost:3000"
    bearer_token = "your-api-token"  # Optional

    headers = {
        "Content-Type": "application/json"
    }

    # Add bearer token if available
    if bearer_token:
        headers["Authorization"] = f"Bearer {bearer_token}"

    async with aiohttp.ClientSession() as session:
        # 1. List available tools
        async with session.post(
            f"{server_url}/tools/list",
            headers=headers
        ) as response:
            tools = await response.json()
            print("Available tools:", [tool['name'] for tool in tools['tools']])

        # 2. Extract markdown from a webpage
        tool_request = {
            "name": "md",
            "arguments": {
                "url": "https://example.com",
                "clean": True
            }
        }

        async with session.post(
            f"{server_url}/tools/call",
            headers=headers,
            json=tool_request
        ) as response:
            result = await response.json()
            print("Markdown content:", result['content'][:200] + "...")

        # 3. Take a screenshot
        screenshot_request = {
            "name": "screenshot",
            "arguments": {
                "url": "https://example.com",
                "full_page": True
            }
        }

        async with session.post(
            f"{server_url}/tools/call",
            headers=headers,
            json=screenshot_request
        ) as response:
            result = await response.json()
            print("Screenshot saved:", result.get('path', 'Screenshot data returned'))

        # 4. Execute JavaScript on a page
        js_request = {
            "name": "execute_js",
            "arguments": {
                "url": "https://example.com",
                "script": "document.title"
            }
        }

        async with session.post(
            f"{server_url}/tools/call",
            headers=headers,
            json=js_request
        ) as response:
            result = await response.json()
            print("Page title:", result['content'])

# Run the example
if __name__ == "__main__":
    # First, start the MCP server in HTTP mode:
    # docker run -p 3000:3000 \
    #   -e CRAWL4AI_ENDPOINT="https://your-crawl4ai-server.com" \
    #   -e CRAWL4AI_BEARER_TOKEN="your-api-token" \
    #   stgmt/crawl4ai-mcp:latest crawl4ai-mcp --http --port 3000

    asyncio.run(test_crawl4ai_mcp())

Installation for Python integration

pip install aiohttp  # For HTTP client

🤝 Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

📄 License

MIT License - see LICENSE file for details.

🔗 Links

• PyPI Package: https://pypi.org/project/crawl4ai-mcp/
• NPM Package: https://www.npmjs.com/package/crawl4ai-mcp-sse-stdio
• Docker Hub: https://hub.docker.com/r/stgmt/crawl4ai-mcp
• GitHub Repository: https://github.com/stgmt/crawl4ai-mcp

---

Made with ❤️ for the AI community