MCP Browser Use Server (mcp-browser-use-server)

mcp-browser-use-server

An MCP server that enables AI agents to control web browsers using browser-use.

Overview

The MCP Browser Use Server provides a robust, production-ready interface for AI agents to interact with web browsers through the Model Context Protocol (MCP). It supports both Server-Sent Events (SSE) and stdio transports, enabling seamless integration with various AI assistants and tools.

Key Features

• Browser Automation: Full browser control through AI agents
• Dual Transport: SSE and stdio protocol support
• VNC Streaming: Real-time browser visualization
• Async Tasks: Non-blocking browser operations
• Docker Support: Containerized deployment with docker-compose
• Persistent Sessions: Long-lived browser sessions with live inspection, tab control and content extraction (see server/session.py)
• Comprehensive Tests: Industry-ready test suite with 95%+ coverage
• Production Ready: Robust error handling and resource management

Quick Start

Prerequisites

Ensure you have the following installed:

• uv - Fast Python package manager
• Playwright - Browser automation library
• mcp-proxy - Required for stdio mode

# uv (Astral) is required for Python workflows.
# For CI (GitHub Actions) prefer the `astral-sh/setup-uv@v5` action which
# installs and configures `uv` on the runner with caching support. Example:

• name: Set up Python and uv

uses: astral-sh/setup-uv@v5 with: enable-cache: true python-version: "3.11"

# Install mcp-proxy globally
uv tool install mcp-proxy

# Update shell to recognize new tools
uv tool update-shell

Environment Configuration

Create a .env file in the project root:

# Required: OpenAI API key for LLM
# Recommended: set `LLM_MODEL` environment variable to your preferred model
# Default: `gpt-5-mini` (you can override with models like `gpt-4-turbo` or `gpt-4o` if available)
OPENAI_API_KEY=your-openai-api-key-here
LLM_MODEL=gpt-5-mini

# Optional: Custom Chrome/Chromium path
CHROME_PATH=/path/to/chrome

# Optional: Patient mode (wait for task completion)
PATIENT=false

# Optional: Logging level
LOG_LEVEL=INFO

# Control headless browser mode used by the server (true = headless, false = visible)
# Default: true
BROWSER_HEADLESS=true

Security Note: Never commit your .env file to version control. It's already included in .gitignore.

Installation

From Source (Development)

# Clone the repository
git clone https://github.com/hubertusgbecker/mcp-browser-use-server.git
cd mcp-browser-use-server

# Install dependencies with uv
uv sync

# Install Playwright browsers
uv pip install playwright
uv run playwright install --with-deps --no-shell chromium

As a Package (Production)

# Build and install from source
uv build
uv tool install dist/mcp_browser_use_server-*.whl

Usage

SSE Mode (Recommended for Web Interfaces)

Server-Sent Events mode is ideal for web-based integrations and remote access.

# Run from source
uv run server --port ${HOST_PORT:-8081}

# Or if installed as tool
mcp-browser-use-server run server --port ${HOST_PORT:-8081}

The server will be available at http://localhost:${HOST_PORT:-8081}/sse

stdio Mode (For Local AI Assistants)

Standard I/O mode is perfect for local AI assistants like Claude Desktop, Cursor, or Windsurf.

# Build and install as global tool
uv build
uv tool uninstall mcp-browser-use-server 2>/dev/null || true
uv tool install dist/mcp_browser_use_server-*.whl

# Run with stdio transport
mcp-browser-use-server run server --port ${HOST_PORT:-8081} --stdio --proxy-port ${PROXY_PORT:-9000}

Note: this package also exposes a CLI alias mcp-browser-use-cli that points to the same entrypoint. You can use either mcp-browser-use-server or mcp-browser-use-cli after installation.

Docker Deployment

Docker provides an isolated, reproducible environment with VNC support for visualization.

# Using docker-compose (recommended)
docker-compose up -d

# View logs
docker-compose logs -f

# Stop services
docker-compose down

# Or use Docker directly
docker build -t mcp-browser-use-server .
docker run --rm -p ${HOST_PORT:-8081}:8081 -p 5900:5900 \
  -e OPENAI_API_KEY=your-key \
  mcp-browser-use-server

Docker Compose Profiles

The docker-compose.yaml supports multiple profiles:

# Development mode with hot reload
docker-compose --profile dev up

# With stdio proxy
docker-compose --profile stdio up

# With monitoring (Prometheus + Grafana)
docker-compose --profile monitoring up

Client Configuration

SSE Mode Configuration

For web-based clients and remote connections:

{
  "mcpServers": {
    "mcp-browser-use-server": {
      "url": "http://localhost:${HOST_PORT:-8081}/sse"
    }
  }
}

stdio Mode Configuration

For local AI assistants:

{
  "mcpServers": {
    "mcp-browser-use-server": {
      "command": "mcp-browser-use-server",
      "args": [
        "run",
        "server",
        "--port",
        "${HOST_PORT:-8081}",
        "--stdio",
        "--proxy-port",
        "9000"
      ],
      "env": {
        "OPENAI_API_KEY": "your-api-key"
      }
    }
  }
}

Configuration Paths by Client

Testing

The project includes a comprehensive test suite for ensuring reliability and correctness.

Running Tests

# Run all tests
./run_tests.sh all

# Run only unit tests
./run_tests.sh unit

# Run integration tests
./run_tests.sh integration

# Run with coverage report
./run_tests.sh coverage

# Run fast tests only (skip slow tests)
./run_tests.sh fast

# Or use pytest directly
uv run pytest tests/
uv run pytest tests/ -v --cov=src --cov=server

CI
--
The repository includes a GitHub Actions workflow that runs ruff and the fast test
suite on pushes and pull requests to `main`. The workflow file is at
`.github/workflows/ci-lint-test.yml`.

Run the same checks locally before pushing:

Format and lint with ruff (via the project's uv wrapper)

uv run ruff format . uv run ruff check .

Run the fast test suite (the same target used by CI)

./run_tests.sh fast

Test Categories

• Unit Tests: Fast, isolated tests for individual functions
• Integration Tests: Test component interactions
• E2E Tests: Full system tests (require RUN_E2E_TESTS=true)
• Performance Tests: Load and performance validation
• MAGG Integration Tests: Shell script for testing Magg aggregator integration

MAGG Integration Testing

The project includes a shell script for validating integration with the Magg aggregator:

# Run MAGG integration test
./tests/test_magg_integration.sh

# The script automatically:
# - Reads HOST_PORT from .env file if present
# - Checks Docker container status and health
# - Verifies Magg configuration
# - Tests browser tools availability through Magg
# - Validates end-to-end browser automation (hubertusbecker.com summary)

Requirements:

• Docker container running on the configured port (default: 8081)
• Magg running and configured (see .magg/config.json)
• The mbro command-line tool available

Port Configuration: The script automatically reads HOST_PORT from .env:

# .env
HOST_PORT=8083  # Match your docker-compose port mapping

If the health check fails, the script provides troubleshooting steps.

CI/CD Integration

Tests are designed to run in CI/CD pipelines. See .github/workflows/ for GitHub Actions examples.

VNC Browser Visualization

Watch browser automation in real-time using VNC:

# Start server with VNC (Docker)
docker-compose up -d

# Connect using noVNC (browser-based)
git clone https://github.com/novnc/noVNC
cd noVNC
./utils/novnc_proxy --vnc localhost:5900

# Or use any VNC client
# Default password: browser-use

Then open in your browser to watch the automation.

Development

Local Development Workflow

# Install development dependencies
uv sync --all-extras

# Run linters and formatters
uv run ruff check .
uv run ruff format .
uv run black .
uv run isort .

# Type checking
uvx ty check .

# Run tests during development
uv run pytest tests/ -v

# Build package
uv build

Project Structure

mcp-browser-use-server/
├── src/
│   └── mcp_browser_use_server/  # Package source
│       ├── __init__.py
│       ├── cli.py               # CLI interface
│       └── server.py            # Server re-exports
├── server/
│   ├── __init__.py
│   ├── __main__.py
│   └── server.py                # Core server implementation
├── tests/                       # Comprehensive test suite
│   ├── conftest.py              # Pytest fixtures
│   ├── test_config.py           # Configuration tests
│   ├── test_browser_tasks.py   # Browser task tests
│   ├── test_mcp_server.py      # MCP server tests
│   ├── test_integration.py     # Integration tests
│   ├── test_e2e.py             # End-to-end tests
│   └── test_performance.py     # Performance tests
├── docker-compose.yaml          # Docker orchestration
├── Dockerfile                   # Container definition
├── pyproject.toml              # Project configuration
├── pytest.ini                  # Test configuration
└── README.md                   # This file

Contributing

Contributions are welcome! Please follow these guidelines:

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Make your changes and add tests
4. Run the test suite (./run_tests.sh all)
5. Commit your changes (git commit -m 'Add amazing feature')
6. Push to the branch (git push origin feature/amazing-feature)
7. Open a Pull Request

Code Quality Standards

• Test Coverage: Maintain >95% coverage
• Type Hints: All functions must have type annotations
• Documentation: Docstrings for all public APIs
• Linting: Code must pass ruff and ty checks
• Formatting: Use black and isort for consistent formatting

Example Usage

Basic Browser Task

Ask your AI assistant:

Navigate to https://news.ycombinator.com and return the top ranked article's title and URL

Advanced Examples

# Search and extract
Go to Google, search for "MCP protocol", and summarize the first 3 results

# Form interaction
Navigate to the contact form at example.com/contact, fill in the fields with test data, and submit

# Data extraction
Visit GitHub's trending page and list the top 5 trending repositories today

# Screenshot capture
Navigate to example.com and take a screenshot of the homepage

API Reference

Available MCP Tools

The server exposes the following tools via MCP:

• run_browser_task: Execute a browser automation task

- Parameters: instruction (string), task_id (optional) - Returns: Task ID for async tracking

• get_task_status: Check the status of a running task

- Parameters: task_id (string) - Returns: Task status, progress, and results

• cancel_task: Cancel a running browser task

- Parameters: task_id (string) - Returns: Cancellation confirmation

• list_all_tasks: List all tasks

- Returns: Array of all tasks with their statuses

New session and browser tools

• browser_get_state: Get current browser state for a task or live session

- Parameters: task_id (optional) or session_id (optional), screenshot (bool) - Returns: JSON state summary; screenshot=true returns base64 PNG

• browser_navigate: Navigate an existing or new live session to a URL

- Parameters: url (string), optional session_id - Returns: Confirmation and updated session state

• browser_click: Click an element in a live session (by index)

- Parameters: session_id, element_index - Returns: Action result

• browser_extract_content: Use session extraction (LLM-assisted) to extract or summarize content

- Parameters: session_id, instruction (string) - Returns: Extracted text or structured result

• Session management: browser_list_sessions, browser_close_session to list and close long-lived sessions

• Tabs API: browser_list_tabs, browser_switch_tab, browser_close_tab for tab control

These tools are implemented in server/server.py and rely on the session manager server/session.py.

Available MCP Resources

• Task Results: Access completed task results via task://{task_id}
• Task History: View task execution history

Troubleshooting

Common Issues

Issue: OPENAI_API_KEY not set

# Solution: Set the API key in .env or environment
export OPENAI_API_KEY=your-key-here

Issue: Chrome/Chromium not found

# Solution: Install Chromium via Playwright
uv run playwright install chromium --with-deps

# Or specify custom path in .env
CHROME_PATH=/usr/bin/chromium-browser

Issue: Tests failing

# Solution: Ensure all dependencies are installed
uv sync --all-extras
uv run playwright install chromium

# Run tests with verbose output
uv run pytest tests/ -vv

Issue: Port already in use

# Solution: Use a different port
uv run server --port 8001

# Or find and kill the process using the port
lsof -ti:8081 | xargs kill -9

Debug Mode

Enable detailed logging for troubleshooting:

# Set log level in .env
LOG_LEVEL=DEBUG

# Or via environment variable
LOG_LEVEL=DEBUG uv run server --port 8081

Architecture

The server follows a modular architecture:

1. MCP Server: Handles protocol communication (SSE/stdio)
2. Browser Manager: Manages browser instances and contexts
3. Task Queue: Async task execution and tracking
4. Agent System: AI-powered browser automation via browser-use
5. Resource Manager: Memory and cleanup management

See AGENTS.md for detailed architecture documentation.

Performance

• Concurrent Tasks: Supports multiple simultaneous browser tasks
• Memory Management: Automatic cleanup of old tasks and browser contexts
• Resource Limits: Configurable via Docker compose
• Async Operations: Non-blocking task execution

Typical performance metrics:

• Task startup: <2 seconds
• Simple navigation: 3-5 seconds
• Complex automation: 10-30 seconds
• Memory per browser: 200-500 MB

Security

Best Practices

• Never commit .env files or API keys to version control
• Use Docker secrets for production deployments
• Limit exposed ports in production
• Set resource limits via Docker to prevent DoS
• Validate all user inputs in browser tasks
• Use HTTPS for SSE mode in production

Environment Isolation

Docker provides strong isolation. For additional security:

# In docker-compose.yaml
security_opt:
  - no-new-privileges:true
read_only: true
tmpfs:
  - /tmp
  - /var/tmp

License

This project is licensed under the MIT License. See LICENSE for details.

Copyright (c) 2025 Dr. Hubertus Becker

Acknowledgments

• Built on browser-use for browser automation
• Uses MCP for AI agent communication
• Powered by Playwright for browser control
• Package management via uv

Support & Contact

• GitHub Issues: Report bugs or request features
• Discussions: Join the conversation
• Email: For private inquiries, contact via GitHub

Changelog

Version 0.9.5 (2025-11-11)

• Complete rebranding and optimization
• Comprehensive test suite added
• Docker Compose support
• Enhanced documentation
• Production-ready deployment configuration

• Repository