A Model Context Protocol (MCP) server implementation for Apache Airflow, enabling seamless integration with MCP clients. This project provides a standardized way to interact with Apache Airflow through the Model Context Protocol.

This project implements a Model Context Protocol server that wraps Apache Airflow's REST API, allowing MCP clients to interact with Airflow in a standardized way. It uses the official Apache Airflow client library to ensure compatibility and maintainability.

This project depends on the official Apache Airflow client library (apache-airflow-client). It will be automatically installed when you install this package.

Set the following environment variables:

AIRFLOW_HOST=<your-airflow-host>        # Optional, defaults to http://localhost:8080
AIRFLOW_API_VERSION=v1                  # Optional, defaults to v1
READ_ONLY=true                          # Optional, enables read-only mode (true/false, defaults to false)

Choose one of the following authentication methods:

Basic Authentication (default):

AIRFLOW_USERNAME=<your-airflow-username>
AIRFLOW_PASSWORD=<your-airflow-password>

JWT Token Authentication:

AIRFLOW_JWT_TOKEN=<your-jwt-token>

To obtain a JWT token, you can use Airflow's authentication endpoint:

ENDPOINT_URL="http://localhost:8080"  # Replace with your Airflow endpoint
curl -X 'POST' \
  "${ENDPOINT_URL}/auth/token" \
  -H 'Content-Type: application/json' \
  -d '{ "username": "<your-username>", "password": "<your-password>" }'

Note: If both JWT token and basic authentication credentials are provided, JWT token takes precedence.

Add to your claude_desktop_config.json:

Basic Authentication:

{
  "mcpServers": {
    "mcp-server-apache-airflow": {
      "command": "uvx",
      "args": ["mcp-server-apache-airflow"],
      "env": {
        "AIRFLOW_HOST": "https://your-airflow-host",
        "AIRFLOW_USERNAME": "your-username",
        "AIRFLOW_PASSWORD": "your-password"
      }
    }
  }
}

JWT Token Authentication:

{
  "mcpServers": {
    "mcp-server-apache-airflow": {
      "command": "uvx",
      "args": ["mcp-server-apache-airflow"],
      "env": {
        "AIRFLOW_HOST": "https://your-airflow-host",
        "AIRFLOW_JWT_TOKEN": "your-jwt-token"
      }
    }
  }
}

For read-only mode (recommended for safety):

Basic Authentication:

{
  "mcpServers": {
    "mcp-server-apache-airflow": {
      "command": "uvx",
      "args": ["mcp-server-apache-airflow"],
      "env": {
        "AIRFLOW_HOST": "https://your-airflow-host",
        "AIRFLOW_USERNAME": "your-username",
        "AIRFLOW_PASSWORD": "your-password",
        "READ_ONLY": "true"
      }
    }
  }
}

JWT Token Authentication:

{
  "mcpServers": {
    "mcp-server-apache-airflow": {
      "command": "uvx",
      "args": ["mcp-server-apache-airflow", "--read-only"],
      "env": {
        "AIRFLOW_HOST": "https://your-airflow-host",
        "AIRFLOW_JWT_TOKEN": "your-jwt-token"
      }
    }
  }
}

Alternative configuration using uv:

Basic Authentication:

{
  "mcpServers": {
    "mcp-server-apache-airflow": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/mcp-server-apache-airflow",
        "run",
        "mcp-server-apache-airflow"
      ],
      "env": {
        "AIRFLOW_HOST": "https://your-airflow-host",
        "AIRFLOW_USERNAME": "your-username",
        "AIRFLOW_PASSWORD": "your-password"
      }
    }
  }
}

JWT Token Authentication:

{
  "mcpServers": {
    "mcp-server-apache-airflow": {
      "command": "uv",
      "args": [
        "--directory",
        "/path/to/mcp-server-apache-airflow",
        "run",
        "mcp-server-apache-airflow"
      ],
      "env": {
        "AIRFLOW_HOST": "https://your-airflow-host",
        "AIRFLOW_JWT_TOKEN": "your-jwt-token"
      }
    }
  }
}

Replace /path/to/mcp-server-apache-airflow with the actual path where you've cloned the repository.

You can select the API groups you want to use by setting the --apis flag.

uv run mcp-server-apache-airflow --apis dag --apis dagrun

The default is to use all APIs.

Allowed values are:

• config
• connections
• dag
• dagrun
• dagstats
• dataset
• eventlog
• importerror
• monitoring
• plugin
• pool
• provider
• taskinstance
• variable
• xcom

You can run the server in read-only mode by using the --read-only flag or by setting the READ_ONLY=true environment variable. This will only expose tools that perform read operations (GET requests) and exclude any tools that create, update, or delete resources.

Using the command-line flag:

uv run mcp-server-apache-airflow --read-only

Using the environment variable:

READ_ONLY=true uv run mcp-server-apache-airflow

In read-only mode, the server will only expose tools like:

• Listing DAGs, DAG runs, tasks, variables, connections, etc.
• Getting details of specific resources
• Reading configurations and monitoring information
• Testing connections (non-destructive)

Write operations like creating, updating, deleting DAGs, variables, connections, triggering DAG runs, etc. will not be available in read-only mode.

You can combine read-only mode with API group selection:

uv run mcp-server-apache-airflow --read-only --apis dag --apis variable

You can also run the server manually:

make run

make run accepts following options:

Options:

• --port: Port to listen on for SSE (default: 8000)
• --transport: Transport type (stdio/sse/http, default: stdio)

Or, you could run the sse server directly, which accepts same parameters:

make run-sse

Also, you could start service directly using uv like in the following command:

uv run src --transport http --port 8080

To install Apache Airflow MCP Server for Claude Desktop automatically via Smithery:

npx -y @smithery/cli install @yangkyeongmo/mcp-server-apache-airflow --client claude

1. Clone the repository:

git clone https://github.com/yangkyeongmo/mcp-server-apache-airflow.git
cd mcp-server-apache-airflow

2. Install development dependencies:

uv sync --dev

3. Create a .env file for environment variables (optional for development):

touch .env

Note: No environment variables are required for running tests. The AIRFLOW_HOST defaults to http://localhost:8080 for development and testing purposes.

The project uses pytest for testing with the following commands available:

# Run all tests
make test

# Run linting
make lint

# Run code formatting
make format

The project includes a GitHub Actions workflow (.github/workflows/test.yml) that automatically:

• Runs tests on Python 3.10, 3.11, and 3.12
• Executes linting checks using ruff
• Runs on every push and pull request to main branch

The CI pipeline ensures code quality and compatibility across supported Python versions before any changes are merged.

Contributions are welcome! Please feel free to submit a Pull Request.

The package is deployed automatically to PyPI when project.version is updated in pyproject.toml. Follow semver for versioning.

Please include version update in the PR in order to apply the changes to core logic.

MIT License