ZenML Integration MCP Server

Overview

The ZenML Integration allows AI assistants to interact with your ZenML MLOps platform through the Model Context Protocol (MCP). This integration provides access to critical information about your machine learning pipelines, runs, artifacts, and services, enabling you to monitor and manage your ML workflows directly through AI interfaces.

Prerequisites

Before setting up the ZenML Integration, you'll need:

1. Access to a ZenML Cloud server (sign up for a free trial at ZenML Cloud if needed)
2. The uv package manager installed locally (install via their installer script or via brew on macOS)
3. A local copy of the ZenML MCP repository

Installation

Step 1: Clone the Repository

First, clone the ZenML MCP repository to your local machine:

git clone https://github.com/zenml-io/mcp-zenml.git

Step 2: Configure Your MCP File

Create an MCP configuration file with the following structure:

{
    "mcpServers": {
        "zenml": {
            "command": "/path/to/uv",
            "args": ["run", "/full/path/to/zenml_server.py"],
            "env": {
                "LOGLEVEL": "INFO",
                "NO_COLOR": "1",
                "PYTHONUNBUFFERED": "1",
                "PYTHONIOENCODING": "UTF-8",
                "ZENML_STORE_URL": "https://your-zenml-server-url.com",
                "ZENML_STORE_API_KEY": "your-api-key-here"
            }
        }
    }
}

Replace the following placeholders:

• /path/to/uv: The full path to your uv executable
• /full/path/to/zenml_server.py: The full path to the zenml_server.py file in the cloned repository
• https://your-zenml-server-url.com: Your ZenML server URL (e.g., https://d534d987a-zenml.cloudinfra.zenml.io)
• your-api-key-here: Your ZenML API key (consider using a service account for this purpose)

Step 3: Set Up with Your AI Assistant

For Claude Desktop:

1. Install Claude Desktop
2. Open Claude Desktop and go to Settings → Developer
3. Click "Edit Config" to open your configuration file
4. Paste your MCP configuration into this file
5. Restart Claude Desktop

For a better experience with tool outputs, go to Settings → Profile and add the following to your preferences:

When using zenml tools which return JSON strings and you're asked a question, you might want to consider using markdown tables to summarize the results or make them easier to view!

For Cursor:

1. Install Cursor
2. In your repository, create a .cursor folder
3. Inside this folder, create a mcp.json file with your MCP configuration
4. Open Cursor settings and enable the ZenML server

Usage

Once configured, you can interact with your ZenML server through your AI assistant. You can ask questions about:

• Users and teams
• Stacks and stack components
• Pipelines and pipeline runs
• Pipeline steps and their logs
• Services and service connectors
• Artifacts and their metadata
• Schedules and pipeline run templates

You can also trigger new pipeline runs if you have run templates configured.

Troubleshooting

• If you encounter connection issues, verify your ZenML server URL and API key
• Ensure the paths to uv and zenml_server.py are correct and absolute
• Check that your ZenML API key has the necessary permissions
• For Cursor, the server might show a red error indicator even when working correctly - test it by asking about your ZenML resources

Feedback and Support

This integration is in beta/experimental release. Join the ZenML Slack community to share your experience and help improve the integration.