loop-browser-automation-mcp

Loop Browser Automation MCP Server

A Model Context Protocol (MCP) server that enables browser automation across multiple AI platforms for coordinated AI workflows.

Overview

This MCP server addresses the limitation that current browser automation tools only work with single applications. It provides advanced browser orchestration to enable true "Loop" functionality across all major AI platforms:

• Claude (claude.ai)
• Perplexity (perplexity.ai)
• ChatGPT (chat.openai.com)
• Gemini (gemini.google.com)

Features

Core Tools

• launch_browser - Launch browser instances for specific AI platforms
• navigate_to_platform - Navigate to AI platform URLs
• send_message_to_ai - Send messages to AI platforms
• get_ai_response - Retrieve responses from AI platforms
• close_browser - Clean up browser sessions
• coordinate_loop - Orchestrate round-robin communication between multiple AI platforms

Browser Support

• Chromium (default) - Best compatibility across platforms
• Firefox - Alternative engine for testing
• WebKit - Safari engine support

Platform Integration

Each AI platform has custom selectors and interaction patterns:

• Claude: Content-editable div inputs with submit buttons
• Perplexity: Textarea with Ask placeholders
• ChatGPT: Textarea with Message placeholders
• Gemini: Rich textarea components

Installation

npm install
npx playwright install

Usage

As MCP Server

Add to your MCP client configuration:

{
  "mcpServers": {
    "loop-browser-automation": {
      "command": "node",
      "args": ["./server.js"]
    }
  }
}

Direct Usage

npm start

Examples

Launch Browser for Claude

{
  "tool": "launch_browser",
  "arguments": {
    "platform": "claude",
    "browserType": "chromium",
    "headless": false
  }
}

Coordinate Multi-Platform Loop

{
  "tool": "coordinate_loop", 
  "arguments": {
    "initialMessage": "What are the key trends in AI development?",
    "platforms": ["claude", "perplexity", "chatgpt"],
    "iterations": 2
  }
}

Architecture

Core Components

• LoopBrowserAutomationServer: Main MCP server class
• Browser Session Management: Track multiple browser instances
• Platform Adapters: Handle platform-specific UI interactions
• Loop Coordination: Orchestrate multi-platform conversations

Session Management

Each browser session is tracked with:

• Unique session ID
• Browser instance
• Page context
• Platform association
• Browser type

Error Handling

• Graceful degradation when platforms change UI
• Session cleanup on errors
• Authentication warnings
• Timeout handling

Security Considerations

• Browsers launch in non-headless mode for authentication
• No credentials are stored or transmitted
• Sessions are isolated per platform
• Automatic cleanup prevents resource leaks

Platform UI Changes

AI platforms frequently update their user interfaces. The server includes:

• Flexible selector strategies
• Error messages indicating UI changes
• Fallback mechanisms
• Easy selector updates

Development

Testing

npm test

Debug Mode

npm run dev

Contributing

Platform selectors may need periodic updates as AI platforms evolve their interfaces. Pull requests welcome for:

• New platform support
• Updated selectors
• Enhanced error handling
• Performance improvements

License

MIT License - See LICENSE file for details

Acknowledgments

Built on the Model Context Protocol by Anthropic and uses Playwright for browser automation.