Browser Operator

🚀 MCP Browser Operator

The MCP Browser Operator is a server designed for browser automation, LLM - controlled browser operations, web page interaction, and content analysis. It is standardized through a JSON - RPC interface.

🚀 Quick Start

The MCP Browser Operator provides a standardized way to automate browser operations, interact with web pages, and perform various tasks. It simplifies the process of browser - related tasks through its well - defined API.

✨ Features

Core Features

1. Browser Management: Create, navigate, and close browser instances.
2. Task Tracking: Asynchronously execute complex tasks such as web page interactions and automation operations.
3. Web Page Interaction: Use LLM to control the browser for advanced operations.
4. Tool Access: Provide debugging tools (e.g., console logs, network logs) and auditing tools (e.g., accessibility audits, performance audits).

Other Features

• Persistent State: Maintain independent browser sessions through project names, supporting multiple projects.
• Security: Include domain blocking, URL validation, and session isolation.

📦 Installation

Requirements

• Python: 3.11+
• Playwright: A browser automation tool
• OpenAI API Key: Required for LLM control

Steps

# Install the MCP operator
pip install mcp-browser-operator

# Install Playwright
playwright install chromium firefox

💻 Usage Examples

Running the Server

# Run in development mode
python -m src/mcp_operator.server dev --port 3000

# Run in production mode
python -m src/mcp_operator.server production --port 3000

Executable Scripts

• run - server: Start the MCP server.
• run - tests: Execute unit and integration tests.
• run - test - harness: Debug using the MCP inspector.

📚 Documentation

Core API Methods

Browser Operations

• Create a browser instance: mcp__browser - operator__createBrowser
• Navigate to a URL: mcp__browser - operator__navigate
• Close the browser: mcp__browser - operator__closeBrowser

Asynchronous Task Management

1. Start a task: Call a browser method to get the job_id.
2. Check progress: Use get - job - status until the task is completed.
3. Process results: Extract results from the task status.

User Notes

• Add a note: mcp__browser - operator__addNote

Debugging Tools

• Get console logs: mcp__browser - tools__getConsoleLogs
• Take a screenshot: mcp__browser - tools__takeScreenshot

Auditing Tools

• Accessibility audit: mcp__browser - tools__runAccessibilityAudit
• Performance audit: mcp__browser - tools__runPerformanceAudit

Project Structure

src/mcp_operator/
├── __init__.py
├── __main__.py
├── server.py
├── browser.py
└── cua/
    ├── agent.py
    ├── computer.py
    └── utils.py

Development Guide

Using the MCP Inspector

# Start by running the test harness
./run - test - harness

# Or use the inspector directly
npx @modelcontextprotocol/inspector ./run - server

Security Measures

• Domain Blocking: Prevent access to malicious websites.
• URL Validation: Validate the security of URLs before navigation.
• Session Isolation: Completely isolate different browser instances.