minicua 0.1.1

A minimal Linux setup for your CUA agent

MiniCUA

A minimal Linux workshop based on Alpine that hosts a custom FastAPI-based endpoint that enables communication from computer using agents (CUAs).

This repository includes:

• A minimal image that includes basic functionalities (browser, terminal, VNC) and an API endpoint
• An MCP-server script that can be connected to the API endpoint

All you need to do is to run the docker image and then plug the tools server into your favorite LLM agent, and you are good to go!

Pulling and running the image

To pull and run the image, simply run:

# Pull the image
docker pull mhz139/minicua

# Run the image
docker run --rm -p 5900:5900 -p 6080:6080 --hostname alpine mhz139/minicua

Since this is equipped with a VNC server, you can go to localhost:6080 to view the server.

In addition, if you have TigerVNC, you can use docker's default address with port 5900 to view it in TigerVNC.

Building the image

If you would like to customize your image, you can clone the repository and then build the Dockerfile directly.

# Clone the repository
git clone https://github.com/MHZ931/MiniCUA.git

# Build the image
docker build -t mhz139/minicua .

Running the server

Before running the server, make sure the image is built and run first.

To run the tools server, the simplest way is to install uv and then use uvx to run the server directly:

# Install uv if you don't have it
curl -LsSf https://astral.sh/uv/install.sh | sh

# Run the server
uvx --from minicua minicua-mcp-server

Or you can add the server into your favorite LLM workflows:

    "minicua-mcp-server": {
      "command": "uvx",
      "args": [
        "--from", "minicua", "minicua-mcp-server"
      ]
    }

Dependencies

• python >= 3.12
• fastmcp>=2.12.4
• pillow>=11.3.0

TODO

• Demo video

Special Thanks to rwildcat/docker_alpine-vnc for offering an awesome barebone VNC connected Linux image to work with.