Astronomer MCP

Astronomer's MCP server puts the Airflow REST API behind MCP tools so an assistant can read and act on the state of a data platform. It lists DAGs with their source code, statistics, warnings and import errors, walks task instances inside a run, and returns task logs directly. Variables, resource pools, assets, installed plugins, provider packages and Airflow configuration are all readable, with connection credentials excluded from what the agent sees. Beyond mirroring endpoints one to one, it ships consolidated tools shaped for agent use: explore_dag gathers everything about a single DAG in one call, diagnose_dag_run assembles the failure context for a broken run, and get_system_health summarizes overall status. It can also trigger runs and pause or unpause DAGs, so an agent that has diagnosed a stuck pipeline can restart it in the same session rather than handing the work back. The server points at any Airflow webserver URL, so it covers self-hosted Airflow as well as Astro deployments. Authentication supports bearer tokens, username and password with automatic OAuth2 exchange on Airflow 3, and basic auth on Airflow 2, with credentials passed as environment variables or CLI flags. Aimed at data engineers who want to debug a failed DAG run from their editor instead of clicking through the Airflow UI.