CoDriver MCP

AI-powered Desktop Automation via Model Context Protocol

CoDriver is an MCP server that gives Claude control over any desktop application. It captures screenshots, reads accessibility trees, performs OCR, and injects mouse/keyboard input - enabling real-time human-AI collaboration on the desktop.

What "Claude in Chrome" is for the browser, CoDriver is for the entire desktop.

Features

Platform Support

Quick Start

Prerequisites

• Node.js 20+
• macOS or Windows 10/11
• Windows only: Visual Studio Build Tools (for robotjs native compilation)

macOS Permissions

CoDriver requires two system permissions. Grant them in System Settings > Privacy & Security:

Tip: If you use CoDriver from an IDE (e.g. Windsurf, VS Code), add the IDE app to both permission lists, then fully restart it (Cmd+Q).

Windows Permissions

Note: robotjs requires Visual Studio Build Tools to compile native modules on Windows. Install via npm install --global windows-build-tools or download from Visual Studio.

Installation

git clone https://github.com/ViktorTrn/codriver-mcp.git
cd codriver-mcp
npm install
npm run build

Configure in Claude Code (Local)

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "codriver": {
      "command": "node",
      "args": ["/absolute/path/to/codriver-mcp/dist/index.js"]
    }
  }
}

Remote Access (HTTP Transport)

node dist/index.js --http                           # localhost:3100
node dist/index.js --http --port 8080               # custom port
node dist/index.js --http --host 0.0.0.0            # all interfaces
node dist/index.js --http --api-key YOUR_SECRET      # with authentication

Remote Claude Code configuration:

{
  "mcpServers": {
    "codriver-remote": {
      "url": "http://your-machine:3100/mcp",
      "headers": {
        "Authorization": "Bearer YOUR_API_KEY"
      }
    }
  }
}

Usage Examples

"Take a screenshot of my desktop"
-> desktop_screenshot

"Read the UI tree of the frontmost app"
-> desktop_read_ui

"Find the Save button"
-> desktop_find { query: "Save" }

"Click the Save button"
-> desktop_click { ref: "ref_3" }

"Drag the file to the trash"
-> desktop_drag { startRef: "ref_5", endRef: "ref_12" }

"Type into the search field"
-> desktop_type { ref: "ref_5", text: "hello world" }

"Launch Safari"
-> desktop_launch { action: "launch", appName: "Safari" }

"Read text from this area of the screen"
-> desktop_ocr { x: 100, y: 200, width: 500, height: 100 }

"Which monitors are connected?"
-> desktop_displays

Accessibility-Driven Workflow

The recommended workflow mirrors Chrome's accessibility approach:

1. Read UI - desktop_read_ui returns an element tree with ref IDs
2. Find elements - desktop_find searches by name, role, or value
3. Interact by ref - desktop_click { ref: "ref_1" } or desktop_type { ref: "ref_3", text: "..." }

This is more reliable than coordinate-based clicking since elements are identified semantically.

Key Combinations

Development

npm run build        # Compile TypeScript
npm run dev          # Watch mode (tsx)
npm test             # Run tests (107 tests)
npm run typecheck    # Type-check without emit

Architecture

CoDriver MCP Server (Node.js/TypeScript)
  |
  +-- Transport
  |     +-- stdio (local, default)
  |     +-- Streamable HTTP/SSE (remote, --http flag)
  |
  +-- Tools (12 total)
  |     +-- desktop_screenshot   PNG/JPEG capture, multi-monitor
  |     +-- desktop_click        Mouse click (coords or ref)
  |     +-- desktop_type         Keyboard input
  |     +-- desktop_key          Key combinations
  |     +-- desktop_scroll       Scroll wheel
  |     +-- desktop_drag         Drag & drop
  |     +-- desktop_windows      Window management
  |     +-- desktop_read_ui      Accessibility tree
  |     +-- desktop_find         Element search
  |     +-- desktop_launch       App lifecycle
  |     +-- desktop_ocr          Text recognition
  |     +-- desktop_displays     Monitor listing
  |
  +-- Modules
        +-- ScreenCapture       screenshot-desktop + sharp (cross-platform)
        +-- InputController     Swift/CGEvent (macOS) | robotjs (Windows)
        +-- WindowManager       Swift/CoreGraphics (macOS) | PowerShell+Win32 (Windows)
        +-- AccessibilityReader JXA/osascript (macOS) | PowerShell+UIA (Windows)
        +-- AppLauncher         AppleScript (macOS) | PowerShell (Windows)
        +-- OcrEngine           tesseract.js (cross-platform)

Tech Stack

Roadmap

• Phase 1: MVP - Screenshots, mouse, keyboard, windows (macOS)
• Phase 2: Accessibility - UI tree reading, element refs, natural language find
• Phase 3: Remote - HTTP/SSE transport, API-key auth, JPEG compression
• Phase 4: Polish - OCR, drag & drop, app launch, multi-monitor
• Phase 5: Windows - Full Windows 10/11 support

Future

• npm package publishing (npm install -g codriver-mcp)
• GIF recording
• Linux platform support

License

MIT

Author

Viktor Trncik - IBT Ingenieurbüro Trncik