Skip to content
MCP server

Airflow MCP Server

MCP server that lets AI clients inspect and control Apache Airflow 3 via its API

Install
Pricing
Open source
Platforms
CLI, self-hosted, API
Runs
Locally or Remotely
Version
0.9.0
Listing updated
today

Similar MCP servers

About Airflow MCP Server

Airflow MCP Server is an open-source MCP server for developer tools. It can run locally on your machine or as a hosted remote server.

The server wraps the Airflow REST API so an AI client can work with DAGs, tasks and connections in an Airflow 3 deployment. Setup needs only a base URL for the Airflow instance and a JWT token, since the server loads the API spec from the instance itself rather than requiring a separate configuration file.

A safe mode restricts it to read-only GET requests, which suits monitoring or troubleshooting sessions where nothing should change. Unsafe mode, the default, exposes write operations as well. By default tools are grouped into hierarchical categories, and a static option exposes everything at once. A resources directory can publish local Markdown guides to the client.

The README is aimed at data engineers and platform teams who run Airflow and want an assistant to query pipeline state or make controlled changes. It supports stdio and Streamable HTTP transports.

Install

Run in a terminal

claude mcp add airflow-mcp-server --env AIRFLOW_BASE_URL='<AIRFLOW_BASE_URL>' --env AUTH_TOKEN='<AUTH_TOKEN>' -- uvx airflow-mcp-server

Add to claude_desktop_config.json

{
  "mcpServers": {
    "airflow-mcp-server": {
      "command": "uvx",
      "args": [
        "airflow-mcp-server"
      ],
      "env": {
        "AIRFLOW_BASE_URL": "<AIRFLOW_BASE_URL>",
        "AUTH_TOKEN": "<AUTH_TOKEN>"
      }
    }
  }
}

Add to ~/.cursor/mcp.json

{
  "mcpServers": {
    "airflow-mcp-server": {
      "command": "uvx",
      "args": [
        "airflow-mcp-server"
      ],
      "env": {
        "AIRFLOW_BASE_URL": "<AIRFLOW_BASE_URL>",
        "AUTH_TOKEN": "<AUTH_TOKEN>"
      }
    }
  }
}

Add to .vscode/mcp.json

{
  "servers": {
    "airflow-mcp-server": {
      "type": "stdio",
      "command": "uvx",
      "args": [
        "airflow-mcp-server"
      ],
      "env": {
        "AIRFLOW_BASE_URL": "<AIRFLOW_BASE_URL>",
        "AUTH_TOKEN": "<AUTH_TOKEN>"
      }
    }
  }
}

Add to ~/.codeium/windsurf/mcp_config.json

{
  "mcpServers": {
    "airflow-mcp-server": {
      "command": "uvx",
      "args": [
        "airflow-mcp-server"
      ],
      "env": {
        "AIRFLOW_BASE_URL": "<AIRFLOW_BASE_URL>",
        "AUTH_TOKEN": "<AUTH_TOKEN>"
      }
    }
  }
}

Add to your client's MCP settings (most use this shape)

{
  "mcpServers": {
    "airflow-mcp-server": {
      "command": "uvx",
      "args": [
        "airflow-mcp-server"
      ],
      "env": {
        "AIRFLOW_BASE_URL": "<AIRFLOW_BASE_URL>",
        "AUTH_TOKEN": "<AUTH_TOKEN>"
      }
    }
  }
}

Using several servers? Build one config file with the MCP config generator.

Configuration

  • AIRFLOW_BASE_URLThe base URL for the Airflow API (e.g., http://localhost:8080)Required
  • AUTH_TOKENThe JWT authentication token for Airflow API accessRequired
  • AIRFLOW_BASE_URLThe base URL for the Airflow API (e.g., http://localhost:8080)Required
  • AUTH_TOKENThe JWT authentication token for Airflow API accessRequired

What it does

Read-only safe mode
A flag restricts the server to GET requests so the assistant cannot modify the Airflow instance.
Hierarchical tool discovery
Tools are grouped into categories such as DAGs, Tasks and Connections, with an option to expose them all statically.
JWT authentication
Connects to Airflow 3 using a JWT token, the only authentication method the README says is supported.
Markdown resources
Points at a folder of Markdown files and exposes each top-level file as a read-only resource.
Multiple transports
Runs over stdio by default or Streamable HTTP for shared, networked deployments.

Good for

  • Checking the status of DAG runs and failed tasks from an AI chat client
  • Reviewing Airflow connections and configuration during troubleshooting
  • Triggering or pausing DAGs with a read-only guard in place
  • Giving an assistant local runbooks as reference resources

Pricing

Open source

Open source, run locally or self-hosted.

Airflow MCP Server FAQ

Which Airflow version does it support?

The README describes it as a server for controlling Airflow 3 and states that only JWT authentication is supported in Airflow 3.0.

Can it run without write access?

Yes. Starting it with the --safe flag limits it to read-only GET operations.

Do I need to include /api/v2 in the base URL?

No. The README says to use the root Airflow URL, and the server fetches the OpenAPI spec from that base.

Is Airflow MCP Server free?

Yes. Airflow MCP Server is open source, so you can use and self-host it for free.

How do I add Airflow MCP Server to Claude Code?

Run this command in your terminal: claude mcp add airflow-mcp-server --env AIRFLOW_BASE_URL='<AIRFLOW_BASE_URL>' --env AUTH_TOKEN='<AUTH_TOKEN>' -- uvx airflow-mcp-server

What are the best alternatives to Airflow MCP Server?

The closest alternatives to Airflow MCP Server are Api Test MCP, Estuary Flow MCP by usefulapi, fastapi_mcp, Diagrid MCP Server and awesome-mcp-servers.

Where this comes from

  • Description: written from the project's README and its MCP Registry entry.
  • Install steps: from the official MCP Registry entry.

How AIPoweredBot checks servers