Apache Airflow MCP Server
Wraps the Airflow REST API so an agent can inspect and trigger DAGs
Add it to your client
Communityuvx mcp-server-apache-airflow
{
"mcpServers": {
"airflow-mcp": {
"command": "uvx",
"args": [
"mcp-server-apache-airflow"
]
}
}
}- Transport
- both
- Runtime
- Python
- Auth
- Required
- License
- MIT
- Stars
- 174
A community project that maps Airflow's REST API onto MCP tools across roughly fourteen groups: DAGs and DAG runs, task instances and logs, variables, connections, pools, datasets, XComs, plugins, providers, import errors and monitoring endpoints. The practical use is triage — find the failed run, read the task log, check the import errors, then trigger a rerun once the cause is understood. A --read-only flag or READ_ONLY=true removes everything that creates, updates or deletes, which is the setting to use against a production scheduler. It points at AIRFLOW_HOST with either basic credentials or a JWT, and speaks stdio, SSE or HTTP depending on the transport flag.
Similar MCP servers
Confluent MCP Server
MCP serverManages Kafka topics, Flink SQL and Schema Registry from an assistant.
Dagster+ MCP Server
MCP serverHosted endpoint for operating a Dagster+ deployment from an assistant
Webflow MCP Server
MCP serverReads and edits Webflow sites, CMS items and pages.

Comments(0)
Sign in to comment