Apache Airflow

You can use the Apache Airflow MCP Server to interact with Airflow through the Model Context Protocol in a standardized way. It exposes Airflow REST API endpoints via an MCP interface so you can manage DAGs, DAG runs, variables, connections, datasets, and more from MCP clients in a consistent, auditable manner.

How to use

Connect your MCP client to the MCP server to manage Airflow resources through the standard MCP endpoints. You can perform read operations such as listing DAGs, DAG runs, variables, and health checks, as well as write operations like creating or updating DAGs, DAG runs, variables, and connections. Use the server in either read-write or read-only mode depending on your security requirements. Select the API groups you want to enable with the --apis flag to tailor the exposed capabilities to your use case.

How to install

Prerequisites include Node tooling for MCP client interactions and the MCP server runtime environment you plan to use (for example, the UVX/UV runtimes or a compatible Node/CLI setup). You will also need access to an Airflow instance and its REST API endpoint.

Step 1: Prepare environment variables and runtime you will use to run the MCP server. Typical variables include the Airflow host, authentication credentials (username/password or a JWT token), and the desired API mode (read-only or read/write). The server defaults to an Airflow host of http://localhost:8080 and API version v1 if you do not override them.

Step 2: Run the MCP server using your preferred launcher. You can run it directly with the MCP runner, using an explicit command and environment configuration. For example, you can start via a runtime command that executes the MCP server module with the appropriate arguments for your environment.

Step 3: (Optional) If you are integrating with Claude Desktop, provide a config snippet that points the MCP server to Airflow and sets authentication. You can use either Basic Authentication (username and password) or JWT Token Authentication. You may also enable read-only mode by setting a flag or environment variable.

Step 4: If you prefer a different runtime (such as uv), ensure your command includes the proper directory and run action, for example, a command that targets the local MCP server package and starts it with the correct arguments.

Additional sections

Configuration and runtime details are designed to be straightforward: you configure the Airflow host, provide credentials, and choose whether to run in read-only mode. The server exposes a comprehensive set of API groups corresponding to DAGs, DAG runs, tasks, variables, connections, pools, datasets, monitoring, health, and more. You can choose which groups to enable to match your operational needs.

Available tools

DAG management

Endpoints to list DAGs, get details, pause/unpause, update, delete, and fetch DAG sources, as well as batch operations and reparse actions.

DAG runs

Endpoints to list, create, get, update, delete, batch retrieve, clear, and annotate DAG runs, including upstream dataset events.

Tasks

Endpoints to list tasks, get task details, access task instances and logs, clear, and update task instance state and tries.

Variables

Endpoints to list, create, read, update, and delete Airflow variables.

Connections

Endpoints to list, create, read, update, delete, and test connections.

Pools

Endpoints to list, create, read, update, and delete pools.

XComs

Endpoints to list and read XCom entries for task instances.

Datasets

Endpoints to list, read datasets and dataset events, and manage queued events.

Monitoring

Health checks and system monitoring endpoints to ensure the MCP server is functioning.

DAG Stats

Endpoint to retrieve DAG statistics.

Config

Endpoint to fetch server configuration details.

Plugins

Endpoint to retrieve installed plugins.

Providers

Endpoint to list available providers.

Event Logs

Endpoints to list and read event logs.

System

Endpoints for import errors and health status.