Universal MCP Generator Framework - The 3D Printer for AI Automation

Transform any website into controllable automation with a single command.

Point OpenWorldAgent at any website and it will:

1. 🔍 Analyze Interface - VisionCraft + YOLO computer vision
2. 🔐 Handle Authentication - Auto-login/signup with user profile
3. 🗺️ Map Elements - Accessibility tree + element detection
4. 🔧 Generate Tools - Create MCP automation tools dynamically
5. 📦 Build Package - Complete NPM package with server
6. ⚙️ Deploy Automatically - Add to Claude Desktop config via AppleScript

Result: Instant automation for ANY website!

OpenWorldAgent
├── server.js                 # Main MCP server
├── core/
│   ├── orchestrator.js       # Pipeline coordinator
│   ├── vision-analyzer.js    # VisionCraft + YOLO 
│   ├── auth-manager.js       # User auth profile
│   ├── interface-mapper.js   # Element mapping
│   ├── tool-generator.js     # Dynamic tool creation
│   ├── npm-packager.js       # Package builder
│   └── config-updater.js     # AppleScript integration
├── credentials/
│   └── profile.json          # User secure profile
└── generated/                # Generated MCP packages

• Environment Variables: All sensitive data stored in .env
• Local Only: Your credentials never leave your machine
• Open Source Safe: No hardcoded personal information
• Gitignore Protected: .env and credentials automatically excluded

1. Clone and Setup:

git clone https://github.com/tylerbessire/OpenWorldAgent.git
cd OpenWorldAgent
npm install

2. Configure Environment:

cp .env.example .env
# Edit .env with your actual credentials

3. Add to Claude Desktop Config:

{
  "mcpServers": {
    "openworldagent": {
      "command": "node",
      "args": ["/path/to/OpenWorldAgent/server.js"],
      "env": {}
    }
  }
}

4. Restart Claude Desktop

5. Run Tests:

npm test

// Point at any website and generate complete automation
generate_automation({
  url: "https://example.com",
  siteName: "example",
  options: {
    skipAuth: false,      // Handle login/signup
    autoDeploy: true      // Auto-add to Claude Desktop
  }
})

• generate_automation - Main tool: URL → Complete MCP package
• analyze_interface - Vision analysis only
• test_authentication - Test login flows
• list_generated_packages - See created packages
• deploy_package - Add package to Claude Desktop
• get_generator_status - System status

Create a .env file with your credentials:

USER_EMAIL=your_email@example.com
USER_PASSWORD=your_secure_password
USER_PHONE=555-555-5555
USER_FIRST_NAME=YourFirst
USER_LAST_NAME=YourLast
USER_ADDRESS=123 Example St, City, State, ZIP

# Optional: Override default paths
MCP_BASE_DIR=/path/to/OpenWorldAgent
CLAUDE_CONFIG_PATH=/path/to/claude_desktop_config.json

Vision Analysis:

• VisionCraft (primary)
• YOLO (fallback)
• Sequential Thinking (complex analysis)

Automation:

• Playwright (browser control)
• AppleScript (macOS integration)
• Accessibility Tree (element mapping)

Integration:

• Auto-config updates
• NPM package generation
• Claude Desktop deployment

Each website generates a complete MCP package:

example-mcp/
├── package.json          # NPM package config
├── server.js            # MCP server with tools
├── README.md            # Usage instructions
└── Auto-deployed to Claude Desktop

generate_automation({
  url: "https://suno.com",
  siteName: "suno"
})
// → Creates suno-mcp package with music generation tools

generate_automation({
  url: "https://github.com",
  siteName: "github"
})
// → Creates github-mcp package with repo management tools

generate_automation({
  url: "https://tradingsite.com",
  siteName: "trading"
})
// → Creates trading-mcp package with buy/sell/analyze tools

OpenWorldAgent: The world's first Universal MCP Generator!

• Point at any website
• Get working automation in minutes
• Control the entire web from Claude Desktop

The "3D Printer for AI Automation" is operational! 🔥

1. Fork the repository
2. Create your feature branch
3. Ensure no credentials are hardcoded
4. Test with .env.example
5. Submit a pull request

MIT License - see LICENSE file for details.

Created by Tyler Bessire - June 2025
The future of AI automation is here.