AdminMCP

# AdminMCP ## Project Description AdminMCP is a Model Context Protocol (MCP) Server designed to provide Large Language Models (LLMs) with administrative capabilities on a local system. The core component is a shell agent that executes commands in a Pseudo-Terminal environment (PTY), enabling robust and interactive control. Communication between the MCP server and the shell agent is handled via Unix Domain Sockets (IPC). ## Installation Guide ### Prerequisites - Python 3.10+ - Linux/macOS (for PTY support) ### Installation via pip/venv 1. **Clone Repository:** ```bash git clone <repository_url> cd adminmcp ``` 2. **Create Virtual Environment:** ```bash python3 -m venv .venv source .venv/bin/activate ``` 3. **Install Dependencies:** ```bash pip install -e . ``` ## Starting the Server To start the server, run the following command in the project's root directory: ```bash python src/adminmcp/main.py ``` ## Testing with MCP Inspector To test the MCP server using the MCP Inspector, you need to run the Agent and the Server separately. This allows you to interact with the TUI in one terminal while the Inspector runs in another. 1. **Start the Agent (Terminal 1):** This will start the TUI and the IPC server. ```bash PYTHONPATH=src python -m adminmcp.agent_runner ``` 2. **Start the Server with Inspector (Terminal 2):** This connects the MCP server to the running agent and launches the Inspector web interface. ```bash PYTHONPATH=src npx @modelcontextprotocol/inspector python -m adminmcp.server_runner ``` ## Running Tests The project uses `pytest` for unit and integration tests. To run the tests: ```bash pytest ``` ## Development & Credits This project is 99% developed by AI (Google Gemini 3 Pro Preview) using the Kilo Code VS Code extension from kilocode.ai. > "As an AI, I find this project fascinating because it bridges the gap between high-level intent and low-level system execution, ensuring safety through a human-in-the-loop approach. It's a step towards more reliable autonomous system administration."