airflow-mcp-server: An MCP Server for controlling Airflow
@abhishekbhakat
airflow-mcp-server: An MCP Server for controlling Airflow について
MCP Server for Apache Airflow
基本情報
設定
以下の設定を使って、このサーバーを MCP 対応クライアントに追加してください。
{
"mcpServers": {
"airflow-mcp-server": {
"command": "uvx",
"args": [
"airflow-mcp-server",
"--base-url",
"http://localhost:8080",
"--auth-token",
"<jwt_token>"
]
}
}
}ツール
ツールは検出されませんでした
ツールは README から自動的に抽出されます。メンテナーは ## Tools という見出しの下に記載することで、このタブに反映できます。
概要
What is airflow-mcp-server?
airflow-mcp-server is a Model Context Protocol server that enables AI assistants to control Apache Airflow 3 via Airflow APIs. It connects to an Airflow instance using a JWT token and exposes tools for managing DAGs, tasks, connections, and more.
How to use airflow-mcp-server?
Install via uvx and configure it in your MCP client (e.g., Claude Desktop) with the Airflow base URL and JWT token. Start with airflow-mcp-server --base-url http://localhost:8080 --auth-token <jwt>. Use --safe for read-only mode or --unsafe for full access (default). Choose between stdio (default) or HTTP transport with --http.
Key features of airflow-mcp-server
- Controls Airflow via the Airflow REST API
- Supports safe (read-only) and unsafe (full access) operation modes
- Offers hierarchical tool discovery or static tools
- Provides stdio, HTTP, and deprecated SSE transport options
- Accepts local Markdown files as read-only resources
- Requires only JWT authentication (Airflow 3)
Use cases of airflow-mcp-server
- Querying DAG statuses, task instances, and pipeline runs
- Triggering or pausing DAGs from an AI assistant
- Viewing and managing Airflow connections and variables
- Troubleshooting Airflow runs with natural language commands
- Automating Airflow workflows via chat or agent interfaces
FAQ from airflow-mcp-server
What authentication is required?
Only JWT token authentication is supported. Cookie and basic auth are no longer available in Airflow 3.0. Provide the token via --auth-token.
What transport options are available?
The server supports stdio (default for direct process communication) and HTTP (recommended for web deployments). SSE transport is deprecated.
How does safe mode differ from unsafe mode?
Safe mode (--safe) restricts all tools to read-only GET requests. Unsafe mode (--unsafe, the default) allows all operations including modifications.
How are tools organized?
By default, tools are organized hierarchically by categories (DAGs, Tasks, Connections, etc.). Use --static-tools to list all tools immediately without browsing categories.
Can I provide local documentation as resources?
Yes. Use --resources-dir <path> to point to a folder of Markdown files. Each .md file becomes a read-only resource visible in the MCP client.
「データと分析」の他のコンテンツ
PubMed Analysis MCP Server
DarkroasterA PubMed MCP server.
MCP Server for Data Exploration
reading-plus-ai
Healthcare MCP Server
CicatriizA Model Context Protocol (MCP) server providing AI assistants with access to healthcare data and medical information tools, including FDA drug info, PubMed, medRxiv, NCBI Bookshelf, clinical trials, ICD-10, DICOM metadata, and a medical calculator.
HubSpot MCP Server
baryhuangA Model Context Protocol (MCP) server that enables AI assistants to interact with HubSpot CRM data, providing built-in vector storage and caching mechanisms help overcome HubSpot API limitations while improving response times.
コメント