Dagster Skills plugin
The official Dagster kit for agents: a reference for assets and dg commands plus access to runs, logs and alerts in the cloud.
- 218 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- The MCP server launches runs and backfills, materializes assets and changes alert policies in a deployed Dagster+ under your user's permissions.
- The `dg launch` and `dg api` commands execute pipelines and call the Dagster+ API with your credentials.
Install
- Works in
Manual install
/plugin marketplace add dagster-io/skillsThen `/plugin install dagster@dagster`; the MCP server ships in the plugin and authorizes through `/mcp`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The MCP server launches runs and backfills, materializes assets and changes alert policies in a deployed Dagster+ under your user's permissions.
- The `dg launch` and `dg api` commands execute pipelines and call the Dagster+ API with your credentials.
This is third-party code. Review the repository files before installing.
What it does
The `dagster` plugin bundles the `dagster-expert` skill and the Dagster+ MCP server. The skill walks an agent through a Dagster project: creating one with `create-dagster`, asset definitions with dependencies, metadata and partitions, schedules, sensors and declarative automation, components and integrations. It insists on reading the reference files first instead of answering from memory, and prefers `dg` commands (`dg list defs`, `dg check`, `dg launch`, `dg dev`) to reading files by hand. The `dagster-plus` MCP server works with a deployed Dagster+ organization under your user's permissions: runs, assets, deployments, code locations, alerts, Issues and metrics. Through it the agent fetches run logs, re-runs failed runs and backfills, materializes assets, and creates and updates alert policies. `dg api` covers the same resources plus the parts the server does not expose.
Who it is for. Data specialists on Dagster who write pipelines with an agent and triage failed runs in Dagster+ without switching to the web UI.
Good fit when
- A project gets its first asset and the agent needs the right `dg` commands and dependencies.
- The choice between a schedule, a sensor and declarative automation is still open.
- A run failed overnight and the agent reads the logs, finds the cause and outlines a fix.
- You want to know which assets lack alert policies, or see the share of successful materializations over a month.
- A new external tool is being connected through Dagster's index of integration libraries.
Not a fit when
- The organization is not on Dagster+: the MCP server works only with it, leaving the skill for local work.
- Pipelines are written in Airflow or Prefect: the reference covers Dagster only.
- The agent must not launch or re-run jobs in a production deployment: the server can start runs and backfills.
Example request
Fetch the logs of the most recent failed run, find the cause and propose a fix.Limitations
MCP needs Dagster+: after installing, run `/mcp` and authorize `dagster-plus`. The default URL points to the US region, so for the EU set `DAGSTER_CLOUD_MCP_URL=https://mcp.agent.eu.dagster.cloud/mcp` before launching Claude Code. The server uses your user's permissions, and alternative authentication methods are described in the Dagster documentation. Installing through `npx skills add` copies only the skills, so the MCP server is not connected that way. Local `dg` commands come with the `dagster-dg-cli` package, usually run through `uv`. Auto-update is off by default for third-party marketplaces.
How to disable. In Claude Code run `/plugin uninstall dagster`: the MCP server goes with the plugin. If you installed through `npx skills`, delete the `dagster-expert` skill folder from the agent's directory.
FAQ
How do I install the Dagster plugin in Claude Code?
Run `/plugin marketplace add dagster-io/skills`, then `/plugin install dagster@dagster` and call the skill as `/dagster-expert`. The Dagster+ MCP server is part of the plugin, and only authorization through `/mcp` remains.
How does the dagster plugin differ from dagster-expert?
The skill is the same, but it now ships in the `dagster` plugin together with the MCP server. The old `dagster-expert` plugin became a stub with no guidance and will be removed; to migrate, uninstall it with `/plugin uninstall dagster-expert`.
What can the Dagster+ MCP server do?
Launch runs, materialize assets, re-run runs and backfills, fetch logs, asset definitions and deployment details, create and update alert policies and read Insights metrics.
How do I connect to the EU region?
Set `export DAGSTER_CLOUD_MCP_URL=https://mcp.agent.eu.dagster.cloud/mcp` before launching Claude Code. By default the URL points to the US region.
How do I install the plugin in Codex and Copilot CLI?
For Codex run `codex plugin marketplace add dagster-io/skills --ref release-stable`, then `codex plugin add dagster@dagster`. For Copilot CLI: `copilot plugin install dagster-io/skills:plugins/dagster`.
Discussion
No questions or reviews yet.
Related
PluginsAn open-source agent for data engineers: a terminal SQL client with warehouse context, a semantic layer, REST and an MCP server.
Astronomer plugin for data engineers: an Airflow MCP server, the af CLI and skills for DAGs, lineage and warehouse queries
An agent queries GreptimeDB metrics, logs and traces with SQL, TQL and RANGE, with write protection and data masking
The Dex plugin: an agent explores a database, builds dbt models and a semantic layer and tracks schema drift under cost control
Databricks Field Engineering's agent kit: an installer for Databricks skills, an MCP server and a visual Builder App
An MCP server and skills from TigerData that help agents write PostgreSQL schemas and queries from current docs and practices
README in short
Every install path delivers the `release-stable` branch, which moves with the weekly Dagster release, and the `dagster` marketplace entry points to it too. Copilot CLI installs the plugin from `master` because it cannot target a branch. Compatibility is stated for Claude Code, Cursor, Codex, Copilot, OpenCode and Pi.