3.6.5

[call518]

MCP-Airflow-API v3.6.5

MCP-Airflow-API is an open-source MCP (Model Context Protocol) server that wraps the Apache Airflow REST API, enabling natural language management of Airflow clusters through LLM-based clients such as Claude Desktop, OpenWebUI, or any MCP-compatible tool.

Overview

Instead of constructing REST API calls manually, you can manage your Airflow workflows using plain language. The server translates natural language tool invocations (via MCP) into the appropriate Airflow REST API requests behind the scenes.

Key Features

Multi-Version Airflow API Support

• API v1 (Airflow 2.x): 43 tools covering full Airflow 2.x cluster management
• API v2 (Airflow 3.x): 45 tools including v2-exclusive asset management tools for data-aware scheduling
• Dynamic version selection via AIRFLOW_API_VERSION environment variable

MCP Tools

The following tool categories are available across both API versions:

• DAG Management: list, get details, pause/unpause DAGs
• DAG Run Management: list and retrieve DAG run information, run durations
• Task Instance Management: list, retrieve details, batch operations, log retrieval, task durations
• Event Log Management: list, filter, summarize DAG event logs
• Import Error Handling: list and retrieve import error details
• Pool Management: list and retrieve pool details
• Variable Management: list and retrieve Airflow variables
• XCom Management: list and retrieve XCom entries
• Connection Management: list and retrieve connections
• System Information: health check, version info, time context
• DAG Analysis: task dependency graph, source code inspection, calendar view
• Asset Management (v2 only): asset listing and retrieval for data-aware scheduling

Transport Modes

• stdio: standard MCP protocol for direct CLI client integration
• streamable-http: HTTP-based transport for web-based MCP clients (e.g., OpenWebUI)

Authentication

• Basic authentication (username/password) for Airflow API
• Bearer token authentication for securing the MCP server endpoint itself
• Remote authentication support for multi-environment setups

Deployment Options

• PyPI: installable via uvx with no local setup required
• Docker Compose: full demo stack including OpenWebUI and MCPO proxy
• Smithery.ai: one-click cloud deployment

Prompt Templates

Built-in prompt template covering common Airflow operational scenarios:

• Cluster health monitoring
• DAG failure troubleshooting
• Task instance analysis
• Capacity and resource management
• Scheduled run verification

Technical Stack

Component	Detail
Runtime	Python 3.12+
MCP Framework	FastMCP
HTTP Client	aiohttp (async, connection pooling, retry strategy)
Packaging	setuptools-scm (version derived from git tags)
CI/CD	GitHub Actions (PyPI publish on tag push), Dependabot
Code Quality	pre-commit, Ruff (formatting/linting), gitleaks (secret detection)

Installation

From PyPI:

uvx --python 3.12 mcp-airflow-api

Docker Compose (full demo stack):

git clone https://github.com/call518/MCP-Airflow-API.git
cd MCP-Airflow-API
cp .env.example .env
# Edit .env with your Airflow credentials
docker-compose up -d

After startup, access OpenWebUI at http://localhost:3002/ and the API docs at http://localhost:8002/docs.

Configuration

Environment Variable	Description	Default
AIRFLOW_API_URL	Base URL of your Airflow instance	required
AIRFLOW_API_USERNAME	Airflow username	required
AIRFLOW_API_PASSWORD	Airflow password	required
AIRFLOW_API_VERSION	API version: v1 or v2	v1
MCP_LOG_LEVEL	Log verbosity	INFO
REMOTE_AUTH_ENABLE	Enable Bearer token auth for MCP server	false

Compatibility

• Apache Airflow 2.x (API v1)
• Apache Airflow 3.x (API v2)
• Python 3.12+

For a test Airflow environment, see the companion project Airflow-Docker-Compose.

---

This discussion was created from the release 3.6.5.