This MCP server provides a Dagster-focused interface for AI agents to discover and interact with Dagster repositories, jobs, assets, and runs. It exposes tools such as list_repositories, list_jobs, list_assets, recent_runs, get_run_info, launch_run, materialize_asset, terminate_run, and get_asset_info. Use these tools to explore your Dagster environment, query pipeline assets, monitor run status, and trigger or materialize assets from an LLM-driven workflow. The server connects to a Dagster GraphQL endpoint over Server-Sent Events (SSE) for streaming updates, enabling real-time feedback during long-running operations.

To use the tools, point your MCP-enabled agent at the Dagster MCP server and issue tool commands like: list_repositories to enumerate available repositories, list_jobs with a repository argument to inspect pipelines, and recent_runs to fetch recent activity. For operational tasks, use launch_run to start a pipeline, then monitor progress via get_run_info or recent_runs. Asset-centric actions like list_assets and materialize_asset let you inspect and materialize Dagster assets, while terminate_run provides a way to stop in-flight executions. The integration is designed for developers and data engineers looking to embed Dagster orchestration capabilities into LLM-assisted workflows.

Prerequisites:

• Python 3.8+ and a working Python environment
• Access to a Dagster instance (GraphQL endpoint) or a local Dagster dev server

Installation (recommended):

1. Install the MCP Dagster server package from PyPI (example package name):

   pipx install mcp-server-dagster

   or, if you prefer pip in a virtual environment:

   python -m pip install mcp-server-dagster

2. Verify installation:

   python -m pip show mcp-server-dagster

3. Run the MCP server (using uvx portable runner):

   uvx mcp-server-dagster

4. If you are not using uvx, you can run via a direct Python entry point or module, depending on distribution specifics:

   python -m mcp_server_dagster

5. Once running, ensure the Dagster GraphQL endpoint is reachable (default: http://localhost:3000/graphql) and that the transport is configured to SSE as described in the configuration.

Note: If you prefer a Docker-based workflow, consult the package documentation for a containerized run option, and adapt the commands accordingly.

• The MCP server assumes Dagster is accessible via the GraphQL endpoint at http://localhost:3000/graphql. Update GRAPHQL_ENDPOINT in the environment if your Dagster instance is elsewhere.
• Transport is SSE by default for streaming updates; ensure your client supports SSE.
• If you run into authentication issues with Dagster, provide the necessary credentials via environment variables or proper Dagster config.
• The available tools map to common Dagster operations; consult Dagster docs for specifics on repository names, job names, and asset identifiers to use with list_repositories, list_jobs, list_assets, and asset-related actions.
• For production deployments, consider configuring proper security, rate limits, and access controls around the MCP endpoints.