We are going to implement an MCP server to orchestrate KitOps for managing and distributing machine learning models. Agents will be able to connect to discover tools for creating, inspecting, pushing, pulling, and removing ModelKits from remote registries like Jozu Hub.

What Makes ModelKits Different?

While Docker containers package applications, ModelKits are purpose-built for AI/ML workflows. They solve the unique challenges AI engineers face when moving projects between environments.

Key Advantages Over Traditional Docker:

Selectively unpack kits — skip pulling what you don’t need
Doubles as your private model registry
One-command deployment

We use:

KitOps for versioning, packaging, and distributing ML models
Jozu Hub as a remote registry for storing and sharing ModelKits
Cursor (MCP Host)

Set Up

Follow these steps one by one:

Install Kit CLI

Here is the documentation for downloading and installing the Kit CLI: Kit CLI Installation for your operating system.

After installing the Kit CLI, you can verify the installation by running:

kit version

Create .env File

Create a .env file in the root directory of your project with the following content:

JOZU_USERNAME=<your_jozu_hub_email>
JOZU_PASSWORD=<your_jozu_hub_account_password>
JOZU_NAMESPACE=<name_of_repository_in_jozu_hub>

All the values are associated with your Jozu Hub account. If you don't have a Jozu account, you can create one at Jozu Hub.

For our demo, we used this Jozu Hub ModelKit: wine-class-prediction. You can also use this ModelKit for your experiments.

Install Dependencies

uv sync

Use MCP Server

Run the MCP server with the created configuration file as mcp.json either globally or in the current project directory. Here's the code of configuring MCP globally to run the server:

{
  "mcpServers": {
    "kitops_mcp": {
      "command": "uv",
      "args": [
        "--directory",
        "/Users/akshay/Eigen/ai-engineering-hub/kitops-mcp",
        "run",
        "--with",
        "mcp",
        "server.py"
      ]
    }
  }
}

Refer to the prompt.txt file for some of the prompts you can use to interact with the MCP server through the MCP host - Cursor in our case.

Get a FREE Data Science eBook 📖 with 150+ essential lessons in Data Science when you subscribe to our newsletter! Stay in the loop with the latest tutorials, insights, and exclusive resources. Subscribe now!

Contribution

Contributions are welcome! Feel free to fork this repository and submit pull requests with your improvements.

README.md Unescape Escape

KitOps MCP Server