mcp-dagster: A Dagster MCP Server

The Model Context Protocol (MCP) is an open protocol that enables seamless integration between LLM applications and external data sources and tools. This repository provides an MCP server for interacting with Dagster, the data orchestration platform.

Overview

A Model Context Protocol server that enables AI agents to interact with Dagster instances, explore data pipelines, monitor runs, and manage assets. It serves as a bridge between LLMs and your data engineering workflows.

Read our launch post to learn more.

Related MCP server: Databricks MCP Server

Components

Tools

The server implements several tools for Dagster interaction:

• list_repositories: Lists all available Dagster repositories

• list_jobs: Lists all jobs in a specific repository

• list_assets: Lists all assets in a specific repository

• recent_runs: Gets recent Dagster runs (default limit: 10)

• get_run_info: Gets detailed information about a specific run

• launch_run: Launches a Dagster job run

• materialize_asset: Materializes a specific Dagster asset

• terminate_run: Terminates an in-progress Dagster run

• get_asset_info: Gets detailed information about a specific asset

Configuration

The server connects to Dagster using these defaults:

• GraphQL endpoint: http://localhost:3000/graphql

• Transport: SSE (Server-Sent Events)

Quickstart

Running the Example

1. Start the Dagster instance with your pipeline:

uv run dagster dev -f ./examples/open-ai-agent/pipeline.py

2. Run the MCP server with SSE transport:

uv run examples/open-ai-agent/run_sse_mcp.py

3. Start the agent loop to interact with Dagster:

uv run ./examples/open-ai-agent/agent.py

Example Interactions

Once the agent is running, you can ask questions like:

• "What assets are available in my Dagster instance and what do they do?"

• "Can you materialize the continent_stats asset and show me the result?"

• "Check the status of recent runs and provide a summary of any failures"

• "Create a new monthly aggregation asset that depends on continent_stats"

The agent will use the MCP server to interact with your Dagster instance and provide answers based on your data pipelines.