Mcp Airflow

mcp-airflow

MCP server that exposes Apache Airflow REST API operations as tools. Built with FastMCP.

Install

# Run directly with uvx (no install needed)
uvx mcp-airflow

# Or install with pip
pip install mcp-airflow

For development:

uv pip install -e ".[dev]"
# or with dependency groups
uv sync --group dev

Configuration

Set these environment variables (or create a .env file from .env.example):

| Variable | Description | Example | |----------|-------------|---------| | AIRFLOW_BASE_URL | Airflow REST API base URL. Use /api/v2 for Airflow 3.x or /api/v1 for 2.x | http://100.x.x.x:8080/api/v2 | | AIRFLOW_USERNAME | Auth username (JWT on 3.x, basic auth on 2.x) | admin | | AIRFLOW_PASSWORD | Auth password | |

Authentication

The client picks the auth scheme automatically based on your Airflow version:

• Airflow 3.x (JWT) — a JWT token is obtained from the /auth/token endpoint

using AIRFLOW_USERNAME/AIRFLOW_PASSWORD, sent as a Bearer token, and refreshed automatically. Point AIRFLOW_BASE_URL at /api/v2.

• Airflow 2.x (basic auth) — if the JWT flow is unavailable, the client falls

back to HTTP basic auth with the same username/password. Point AIRFLOW_BASE_URL at /api/v1.

Usage

Run the server:

mcp-airflow

Or add to your MCP client config (e.g., Claude Desktop):

{
  "mcpServers": {
    "airflow": {
      "command": "mcp-airflow",
      "env": {
        "AIRFLOW_BASE_URL": "http://100.x.x.x:8080/api/v2",
        "AIRFLOW_USERNAME": "admin",
        "AIRFLOW_PASSWORD": "your-password"
      }
    }
  }
}

Tools

| Tool | Description | |------|-------------| | list_dags | List all DAGs with paused/active status | | get_dag_runs_today | Get all DAG runs from today with status | | get_dag_run_status | Get the latest run status for a specific DAG | | trigger_dag_run | Trigger a manual DAG run | | get_task_instances | Get task instances for a specific DAG run | | check_failed_dags | Check for failed DAGs in the last 24 hours | | check_scheduler_health | Check scheduler heartbeat and metadatabase status |

Tests

pytest

License

MIT

Source & license

This open-source MCP server is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: antonio-mello-ai
• Source: antonio-mello-ai/mcp-airflow
• License: MIT

Install and usage instructions live in the source repository linked above.