MCP Endpoints for Apache Airflow

The Apache Airflow MCP endpoint enables AI clients that support the Model Context Protocol (MCP) to view, start, troubleshoot and create Airflow DAGs on your Data Fabric cluster, with every action running as the user who asked for it.

The Apache Airflow MCP enables AI clients that support the Model Context Protocol (MCP) to view, start, troubleshoot and create Apache Airflow DAGs on the Data Fabric cluster. The endpoint is available at https://<host>:5679/airflow. It is turned off by default.

To configure the Airflow endpoint, add an Airflow section to the configuration file (/opt/mapr/data-access-gateway/conf/mcp-server.conf). These are the parameters you can set:

  • airflow.enabled: turns on the Airflow endpoint (false by default)

  • mcp.airflow.rest-service-url: the Airflow REST API URL (found automatically by default)

  • airflow.auth.use-internal-jwt: uses Airflow's own username and password instead of passing the user's JWT. For development and testing only (false by default)

  • airflow.auth.username: the Airflow username to use when use-internal-jwt is true

  • airflow.auth.password: the Airflow password to use when use-internal-jwt is true

  • mcp.airflow.skip-run-as-user-check: turns off the run_as_user check. For development and testing only (false by default)

  • mcp.airflow.dag-upload-max-size-bytes: the largest DAG file that can be uploaded (10,485,760 bytes, or 10 MB, by default)

  • mcp.airflow.dag-upload-ttl: how long a review token lasts (30m by default)

  • mcp.airflow.dag-stage-file-permissions: permissions on staged DAG files (rw------- by default)

  • mcp.airflow.dag-review-temp-dir: the folder for staged DAG files (java.io.tmpdir/airflow-dag-review by default)

    Example:
    airflow {
  enabled = true
  rest-service-url = "https://<airflow_rest_api_hostname>:<port>"
}

To use the Airflow endpoint, you need the mapr-data-access-gateway package and Airflow from DEP 10.2.0 or later.

The Airflow endpoint has 33 tools for the Airflow environment, assets, DAGs, DAG runs, tasks, diagnostics and DAG creation.