Can AI agents automate Dagster?

Dagster is automatable today through the dg CLI for a local project, with GraphQL and a beta Dagster+ MCP server beside it. This record is Dagster, including the open-source core and Dagster+. The homepage states that the core orchestration framework is open source and that Dagster+ is the managed cloud offering, including hybrid deployment where compute stays in the customer environment. The public GitHub API reports Apache-2.0 for dagster-io/dagster. The GraphQL API is served at /graphql by the Dagster webserver. Dagster states that the API is still evolving and that a large portion is primarily for the webserver. Documented queries and mutations list runs, repositories, and jobs, launch a run, and terminate a run. For Dagster+, DagsterGraphQLClient sends a user token in the Dagster-Cloud-Api-Token header to a deployment URL. That Python client currently exposes submit_job_execution, get_run_status, reload_repository_location, and shutdown_repository_location. The dg CLI is installed with the dagster-dg-cli package. dg launch starts a run from a local project, dg list defs can print JSON, and dg check defs exits 1 when validation fails. The hosted MCP server is marked Dagster+ only and beta. The US URL is https://mcp.agent.dagster.cloud/mcp and the EU URL is https://mcp.agent.eu.dagster.cloud/mcp. Auth is OAuth, or an Authorization bearer user token plus a Dagster-Cloud-Organization header. The tool table covers runs, run logs, assets, deployments, alert policies, and Dagster+ issues, with launch, update, and terminate on some of those objects. https://docs.dagster.io/llms.txt is a documentation index. iPaaS connector catalogs were not opened. Computer-use viability is unassessed. CLI scores 7. API and MCP score 6. CLI is the best path because dg launch and dg list are maintained commands for a project checkout, while the GraphQL API carries an explicit breaking-change warning and the MCP server is a Dagster+ beta.

Best path today: cli · Overall automatability: 7/10