Dagster FAQs

Question 1

How does this Dagster MCP project differ from the official Dagster MCP?

Accepted Answer

This project is an 'operations-time' tool, focused on monitoring and controlling a running Dagster instance. The official Dagster MCP (`dg[mcp]`) is a 'development-time' tool, designed to help AI write Dagster code and scaffold components. They serve different, complementary purposes.

Question 2

Are the write operations (like launching jobs) enabled by default?

Accepted Answer

No, operational write actions such as launching jobs, terminating runs, and reloading code locations are disabled by default to ensure read-only access. You can explicitly enable them by setting the `DAGSTER_READ_ONLY` environment variable to `false`.

Question 3

What kind of operational actions can I perform with Dagster MCP?

Accepted Answer

You can monitor and inspect Dagster runs (logs, status, step stats), search assets, list jobs, schedules, sensors, and code locations. Optionally, you can also execute actions like launching jobs, terminating runs, and reloading code locations, which are disabled by default for safety.

Question 4

Is Dagster MCP compatible with both Dagster Cloud and self-hosted Dagster instances?

Accepted Answer

Yes, Dagster MCP is designed for compatibility with both self-hosted Dagster instances and Dagster Cloud. Configuration involves setting the `DAGSTER_URL` and, for Dagster Cloud, providing a `DAGSTER_API_TOKEN`.

Question 5

What is Dagster MCP and its primary purpose?

Accepted Answer

Dagster MCP is an MCP (Model Context Protocol) server that wraps the Dagster GraphQL API. Its primary purpose is to provide any MCP client (like Claude Code or Cursor) with full visibility and operational control over your Dagster instances, enabling monitoring and management of runs, assets, jobs, and more.

Dagster

Dagster

主な機能

ユースケース

主な機能

ユースケース