mcp-zenml

This MCP server exposes a ZenML-centric set of MCP tools that enable programmatic access to ZenML resources and actions. You can query core entities like users, projects, stacks, models, and deployments, as well as interact with pipelines and executions. The toolset includes capabilities to list snapshots or deployments, retrieve detailed step logs, trigger pipeline runs via snapshots, and inspect models, artifacts, and service endpoints. By connecting an MCP client (such as an IDE integration or CLI client) to the ZenML MCP server, you can explore your ZenML environment, discover runnable pipelines, and trigger runs in a controlled, auditable way. The available tools are organized into core entities, pipeline execution, organization, and deployment-related groups, plus interactive apps and analysis utilities. Use the snapshot-based workflow to trigger reproducible runs and monitor deployments via the provided status and logs endpoints.

Prerequisites:

• Docker installed on your machine (recommended for quick setup)
• Git installed for cloning the repository (optional if using a prebuilt image)
• Access token or credentials if your ZenML instance requires authentication

Install steps (Docker-based setup):

1. Pull and run the MCP ZenML image:

docker pull zenml/mcp-zenml

docker run -i -p 8080:8080 zenml/mcp-zenml

2. Configure environment variables as needed (examples):

• ZENML_MCP_TOKEN: your access token
• ZENML_API_URL: URL to your ZenML API endpoint (if separate from the MCP container)

3. Verify the server is up by connecting with an MCP client or curling the health endpoint (if exposed by the image).

If you prefer to run locally from source, clone the repository and install dependencies according to the project’s setup instructions, then start the server using the recommended local run command provided in the repository (e.g., python -m mcp_server or a node/uvx command if applicable).

Tips and caveats:

• The ZenML MCP server supports both snapshot-based and deprecated template-based triggers; prefer using snapshots for reproducible runs.
• When running behind a proxy or requiring authentication, set ZENML_API_URL and ZENML_MCP_TOKEN accordingly.
• If you encounter connection issues, ensure the MCP server container has network access to your ZenML backend and that the API endpoint is reachable.
• The available tools are grouped by category (Pipeline Execution, Organization, Core Entities, Interactive Apps, Analysis Tools). Use get_* and list_* tools to navigate resources, then use trigger_pipeline with snapshot_name_or_id to start runs.
• For production deployments, consider persistent storage, proper credential management, and configuring appropriate RBAC/permissions within ZenML and your MCP client.