A Model Context Protocol (MCP) server that provides comprehensive browser automation capabilities using Playwright. This server enables AI assistants to interact with web pages through standardized MCP tools for navigation, content extraction, form filling, and screenshot capture.

• Navigate to URLs with smart waiting strategies
• Extract page content with customizable selectors
• Take screenshots (full page, viewport, or specific elements)
• Execute JavaScript with result capture
• Click elements by CSS selectors
• Fill forms automatically with validation

• Multi-browser support (Chromium, Firefox, WebKit)
• Request interception and monitoring
• Viewport customization and responsive testing
• Link extraction and URL processing
• Error handling with detailed responses
• Resource management and cleanup

• Python 3.8 or higher
• Node.js (for Playwright browser installation)

# Clone the repository
git clone <repository-url>
cd claude-browser-mcp

# Install dependencies
pip install -e .

# Install Playwright browsers
playwright install

pip install claude-browser-mcp
playwright install

Start the server with stdio transport:

browser-mcp
# or
python -m src.server

Configure the browser through environment variables:

export BROWSER_HEADLESS=true          # Run in headless mode
export BROWSER_TYPE=chromium          # Browser type (chromium/firefox/webkit)
export BROWSER_TIMEOUT=30000          # Default timeout in milliseconds

Add to your MCP client configuration:

{
  "mcpServers": {
    "browser-automation": {
      "command": "browser-mcp",
      "args": []
    }
  }
}

Navigate to a specified URL with optional waiting.

{
  "name": "navigate_to",
  "arguments": {
    "url": "https://example.com",
    "wait_for": "selector", 
    "timeout": 30
  }
}

Extract text content from the current page.

{
  "name": "get_page_content", 
  "arguments": {
    "include_links": true,
    "selector": ".main-content"
  }
}

Click on elements by CSS selector.

{
  "name": "click_element",
  "arguments": {
    "selector": "button#submit",
    "timeout": 10
  }
}

Fill form fields with data.

{
  "name": "fill_form",
  "arguments": {
    "fields": {
      "#email": "user@example.com",
      "#password": "secretpass"
    },
    "submit": true
  }
}

Capture page screenshots.

{
  "name": "take_screenshot",
  "arguments": {
    "full_page": true,
    "selector": ".dashboard"
  }
}

Run JavaScript in the browser context.

{
  "name": "execute_javascript", 
  "arguments": {
    "code": "document.title",
    "return_value": true
  }
}

claude-browser-mcp/
├── src/
│   ├── __init__.py          # Package initialization
│   ├── server.py            # MCP server implementation
│   ├── browser.py           # Browser management
│   ├── actions.py           # High-level browser actions
│   └── utils.py             # Utility functions
├── requirements.txt         # Python dependencies
├── setup.py                 # Package configuration
└── README.md               # This file

• MCP server implementation with tool registration
• Request routing and response formatting
• Error handling and logging
• Async tool execution

• Playwright browser lifecycle management
• Context creation and configuration
• Resource cleanup and recovery
• Multi-browser support

• High-level browser automation methods
• Content extraction and processing
• Form interaction and validation
• Screenshot and JavaScript execution

• HTML sanitization and cleaning
• URL validation and normalization
• Image processing and encoding
• Data formatting utilities

• HTML sanitization removes dangerous scripts and attributes
• URL validation prevents malicious redirects
• Input validation for all user-provided data
• Resource limits prevent excessive memory usage
• Timeout controls prevent hanging operations

# Build and run with Docker Compose
docker-compose up browser-mcp

# Or build manually
./scripts/docker-build.sh
./scripts/start-container.sh

# Build production image
docker build -t browser-mcp:latest .

# Run with optimal settings
docker run -d \
  --name browser-mcp \
  --init --ipc=host --shm-size=1gb \
  --memory=2g --cpus=1.0 \
  -v $(pwd)/screenshots:/app/screenshots \
  -v $(pwd)/downloads:/app/downloads \
  browser-mcp:latest

# Development container with debugging
docker-compose --profile dev up browser-mcp-dev

# Access container
docker exec -it claude-browser-mcp-dev /bin/bash

# Health check
./scripts/health-check.sh

# View logs
docker logs -f claude-browser-mcp

# Monitor resources
docker stats claude-browser-mcp

The server provides detailed error responses with:

• Error categorization (timeout, validation, execution)
• Context information (URL, selector, arguments)
• Recovery suggestions where applicable
• Logging for debugging and monitoring

All tools return standardized JSON responses:

{
  "success": true,
  "url": "https://example.com",
  "title": "Page Title",
  "data": "...",
  "metadata": {
    "timestamp": "...",
    "execution_time": "..."
  }
}

Error responses include:

{
  "success": false,
  "error": "Detailed error message",
  "tool": "tool_name",
  "arguments": {...},
  "timestamp": "..."
}

# Install in development mode
pip install -e .[dev]

# Run tests
pytest tests/

# Format code
black src/

# Type checking
mypy src/

1. Define tool schema in server.py
2. Implement action method in actions.py
3. Add utility functions in utils.py
4. Update documentation and tests

MIT License - see LICENSE file for details.

• Playwright for browser automation
• MCP for the protocol specification
• Anthropic for Claude and MCP development

• Issues: Report bugs and request features on GitHub
• Documentation: See inline code documentation
• Community: Join MCP community discussions

Note: This is a foundational implementation. Additional features like request interception, advanced form handling, and performance optimizations can be added based on specific use cases.