MCP‑Airflow‑API: Natural‑Language Control for Apache Airflow

MCP‑Airflow‑API turns the RESTful interface of Apache Airflow into a conversational toolset that AI assistants can call directly. By exposing every endpoint as a natural‑language “tool” under the Model Context Protocol, developers can instruct an LLM to list running DAGs, trigger jobs, or manage assets without writing HTTP requests. This eliminates the need to remember complex URLs, authentication headers, or query parameters, making Airflow management as simple as chatting with a colleague.

The server supports both legacy Airflow 2.x (API v1) and the newest 3.x releases (API v2). A single deployment dynamically loads the appropriate set of tools based on an environment variable, so the same MCP client can work against either version. The v1 set contains 43 tools covering DAG, task, and connection operations, while the v2 release adds two asset‑management tools that unlock data‑aware scheduling features introduced in Airflow 3.0+. The tool descriptions are generated from the official API documentation, ensuring that every command remains up‑to‑date with Airflow’s evolving feature set.

Key capabilities include:

• Intuitive command generation – An LLM can translate “Show me the currently running DAGs” into a valid API call, automatically handling pagination and authentication.
• Dynamic version selection – A single MCP server can serve both Airflow 2.x and 3.x clusters, reducing operational overhead.
• Extensive tool coverage – With 45 tools in total, developers can perform nearly any routine Airflow operation—triggering DAGs, pausing schedules, inspecting logs, and managing assets—all through a unified natural‑language interface.
• Seamless integration – The MCP server exposes standard endpoints (, ) that plug directly into OpenWebUI, Claude, or any other LLM platform that supports MCP.

Real‑world scenarios benefit from this abstraction: a data engineering team can let an AI assistant auto‑trigger nightly runs, generate SLA reports, or troubleshoot failures by simply describing the desired outcome. A DevOps engineer can ask the assistant to update connection strings or rotate secrets, and the MCP server will translate that into the correct Airflow API calls. Because the toolset is generated from live documentation, new Airflow features become immediately available to AI clients without manual updates.

In short, MCP‑Airflow‑API removes the friction between human intent and Airflow’s technical interface. By packaging REST operations as natural‑language tools, it empowers developers to build smarter, more conversational data pipelines and orchestration workflows that adapt effortlessly to evolving Airflow versions.