Playwright Universal MCP

A universal Model Context Protocol (MCP) server for browser automation that works reliably in containerized environments with limited privileges. It provides browser control capabilities to AI assistants like Claude, GPT-4, etc.

• 🌐 Multi-browser support: Choose between Chromium, Chrome, Microsoft Edge, Firefox, or WebKit
• 🐳 Container-friendly: Works in environments with limited privileges (like Docker containers)
• 👁️ Headless/Headful modes: Run in headless mode for server environments or headful mode for debugging
• 🛠️ Extensive browser control: Navigate, click, type, take screenshots, and more
• 📄 Multiple page support: Create and manage multiple browser pages/tabs

# Install the MCP Python SDK
pip install git+https://github.com/microsoft/mcp-python-sdk.git

# Install the MCP server globally
pipx install playwright-universal-mcp

# Install the required browsers
playwright install chromium
# Optional: install other browsers
playwright install firefox webkit msedge chrome

# Create and activate a virtual environment
python -m venv playwright-mcp-venv
source playwright-mcp-venv/bin/activate

# Install the MCP Python SDK
pip install git+https://github.com/microsoft/mcp-python-sdk.git

# Install the package
pip install playwright-universal-mcp

# Install browsers
playwright install chromium

# Clone the repository
git clone https://github.com/xkiranj/playwright-universal-mcp.git
cd playwright-universal-mcp

# Install the MCP Python SDK
pip install git+https://github.com/microsoft/mcp-python-sdk.git

# Install the package
pip install -e .

# Install browsers
playwright install chromium

playwright-universal-mcp --help

usage: playwright-universal-mcp [-h] [--browser {chromium,firefox,webkit,msedge,chrome}] [--headless] [--headful] [--debug] [--browser-arg BROWSER_ARG]

Universal Playwright MCP server with multi-browser support

options:
  -h, --help            show this help message and exit
  --browser {chromium,firefox,webkit,msedge,chrome}, -b {chromium,firefox,webkit,msedge,chrome}
                        Browser to use (default: chromium)
  --headless            Run browser in headless mode (default: true)
  --headful             Run browser in headful mode (with GUI)
  --debug               Enable debug logging
  --browser-arg BROWSER_ARG
                        Additional browser arguments (can be specified multiple times)

# Start with default options (Chromium in headless mode)
playwright-universal-mcp

# Use Microsoft Edge
playwright-universal-mcp --browser msedge

# Use Firefox in headful mode (visible browser window)
playwright-universal-mcp --browser firefox --headful

# Enable debug logging
playwright-universal-mcp --debug

# Pass additional arguments to the browser
playwright-universal-mcp --browser-arg="--disable-gpu" --browser-arg="--window-size=1920,1080"

To use this MCP server with Claude Desktop or other MCP-enabled applications, add the following configuration:

Add the following to your ~/.config/Claude/claude_desktop_config.json file:

{
  "mcpServers": {
    "browser": {
      "command": "playwright-universal-mcp",
      "args": ["--browser", "msedge", "--headless"]
    }
  }
}

To run the MCP server as a persistent service with PM2:

1. Create a PM2 configuration file:

cat > ~/playwright-universal-mcp.config.js << 'EOF'
module.exports = {
  apps: [{
    name: "playwright-universal-mcp",
    script: "playwright-universal-mcp",
    args: "--browser msedge --headless",
    watch: false,
    autorestart: true,
    restart_delay: 3000
  }]
}
EOF

1. Start the service:

pm2 start ~/playwright-universal-mcp.config.js

1. Save the PM2 process list:

pm2 save

This MCP server works well in containerized environments. Here's a simple Dockerfile example:

FROM python:3.10-slim

# Install dependencies
RUN apt-get update && apt-get install -y \
    wget \
    gnupg \
    && rm -rf /var/lib/apt/lists/*

# Install the package
RUN pip install playwright-universal-mcp

# Install browsers (install the ones you need)
RUN playwright install chromium

# Run the MCP server
ENTRYPOINT ["playwright-universal-mcp", "--headless"]

The MCP server provides the following tools:

• navigate: Navigate to a URL
• click: Click on an element by selector or text
• type: Type text into an input element
• get_text: Get text content from an element
• get_page_content: Get the current page HTML content
• take_screenshot: Take a screenshot of the current page
• new_page: Create a new browser page
• switch_page: Switch to a different browser page
• get_pages: List all available browser pages
• wait_for_selector: Wait for an element to be visible on the page
• get_browser_info: Get information about the current browser session

MIT License

This project builds upon:

• Playwright for browser automation
• Model Context Protocol (MCP) for the connection standard
• MCP Python SDK for the MCP implementation