MCP Client

MCP Client

An interactive AI assistant interface with a dual-display system: a terminal-based CLI for commands and an always-on-top window for real-time model output monitoring.

Features

• Terminal Interface: Interact with the AI assistant through a command-line interface
• Always-on-top Display Window: Monitor model outputs in real-time
  • Translucent window with modern design
  • Stays on top of other applications
  • Follows you across virtual desktops
• Real-time Output Monitoring: See what the model is generating as it happens
• Pause Functionality: Pause model execution if you see undesired actions being generated
• User Input Requests: The display window can request user input while long-running tasks continue in the CLI
• Global Shortcuts:
  • Ctrl/Cmd + Shift + T to toggle the follow desktop behavior
  • Ctrl/Cmd + Shift + P to pause/resume model execution
• MCP Integration: Connect to Model Context Protocol (MCP) servers for enhanced AI capabilities

Use Cases (all of them TODO...)

• Computer Control: Use the MCP client to control your computer with AI assistance
• Browser Automation: Navigate and interact with web browsers while monitoring AI actions
• Long-running Tasks: Monitor progress and provide input while background tasks run
• Safety Monitoring: Observe and control AI actions in real-time

Getting Started

Prerequisites

• Node.js (v12 or higher)
• npm
• Python 3.7+
• MCP library (for MCP server integration)

Installation

1. Clone this repository
2. Install dependencies:

   npm install
   pip install -r backend/requirements.txt
   pip install mcp
   

3. Start the application:

   python backend/cli.py
   

Usage

The application runs in two parts:

1. Terminal Interface: Where you input commands and see primary responses
2. Floating Window: Where you can monitor model outputs in real-time

Terminal Commands

The CLI supports various commands for interacting with the AI assistant. Type help in the terminal for a list of available commands.

Display Window Controls

• Pause Button: Click to pause model execution if you see undesired actions
• Toggle Follow Desktop: Enable/disable the window following you across virtual desktops
• Drag: Click and drag anywhere on the window to reposition it

MCP Server Integration

The application can connect to MCP servers for enhanced AI capabilities:

1. Configure your MCP server in backend/server.config:

   {
     "llm_type": "mcp",
     "server_path": "/path/to/your/mcp_server.py",
     "server_name": "your-mcp-server",
     "server_version": "0.1.0"
   }
   

2. Install the MCP library:

   pip install mcp
   

3. Start the application with MCP integration:

   python backend/cli.py
   

The application will automatically connect to the specified MCP server and use it for generating responses. If the MCP server is not available, it will fall back to the mock LLM implementation.

Architecture

The application consists of:

• Electron frontend for the always-on-top display window
• Python backend with Flask API for handling AI interactions
• Terminal-based CLI for user input and command processing

License

ISC