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:

   git clone <repository_url> cd adminmcp

2. Create Virtual Environment:

   python3 -m venv .venv source .venv/bin/activate

3. Install Dependencies:

   pip install -e .

Starting the Server

To start the server, run the following command in the project's root directory:

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.

   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.

   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:

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."