Creativity · MCP — server

MCP Dagster Server

The MCP Dagster Server wraps Dagster's GraphQL API for Dagster Cloud and OSS. MCP clients can list software-defined assets, inspect runs, launch backfills, and read logs — making it easier for analytics engineers to triage pipeline issues conversationally and document lineage on the fly.

MCP facts

Kind: server
Ecosystem: anthropic-mcp
Language: Python
Transports: stdio

Capabilities

Tools: list_assets, get_asset_materializations, launch_backfill, list_runs, fetch_logs
Resources: dagster://asset/{key}, dagster://run/{id}
Auth: Dagster+ token or OSS GraphQL endpoint

Install

pipx install mcp-server-dagster

Configuration

{
  "mcpServers": {
    "dagster": {
      "command": "mcp-server-dagster",
      "env": {
        "DAGSTER_URL": "https://acme.dagster.cloud/prod/graphql",
        "DAGSTER_TOKEN": "${DAGSTER_TOKEN}"
      }
    }
  }
}

Frequently asked questions

What's the advantage of exposing the asset graph?

Software-defined assets encode lineage. An LLM can walk upstream to find the root cause of a failure and downstream to predict who will be affected.

Does it work with Dagster+?

Yes — point DAGSTER_URL at the Dagster+ deployment's GraphQL endpoint and use a user or agent token.

Can it modify pipeline code?

No — MCP exposes runtime operations. Code changes still go through a normal git + CI workflow.

Sources

Dagster — GraphQL API — accessed 2026-04-20
Model Context Protocol — accessed 2026-04-20
Dagster OSS — accessed 2026-04-20