astro-airflow-mcp

What it is and what it does

Astro-airflow-mcp is a Model Context Protocol server that bridges Apache Airflow and AI assistants (Claude, Gemini, Codex, or custom MCP clients). It exposes Airflow's REST API through structured MCP tools, allowing AI systems to query DAG metadata, trigger runs, inspect task logs, manage pools and variables, and diagnose failures—all without direct Airflow access.

The package runs in two modes: as a standalone HTTP or stdio server, or embedded as an Airflow plugin (2.x Flask or 3.x FastAPI). It automatically detects Airflow version and adapts its API calls accordingly. Authentication supports bearer tokens, username/password with OAuth2 exchange on Airflow 3.x, and custom CA certificates. Core dependencies are fastmcp (the MCP framework), pydantic (validation), httpx (HTTP client), typer (CLI), and utilities like pyyaml and simple-term-menu for configuration and interactive selection.

Use it for:

• Enable Claude or other AI assistants to diagnose and troubleshoot failed DAG runs by fetching task logs and instance details
• Automate DAG triggering and monitoring from AI workflows without exposing raw Airflow credentials to the AI system
• Provide read-only Airflow visibility to AI assistants for health checks and system overview without write permissions
• Integrate Airflow management into IDE-based AI coding assistants (Cursor, VS Code) for workflow development and testing
• Query DAG source code, task dependencies, and asset lineage through AI natural-language queries

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Provides AI assistants with programmatic access to Apache Airflow via the Model Context Protocol, exposing DAG management, task execution, monitoring, and configuration tools through a FastMCP server.

Yes, if you run Airflow and want AI assistants to interact with it safely. The package is actively maintained, has no known vulnerabilities, requires only Python 3.10+, and installs with low friction. The unclear license status is a minor concern—verify the actual license before deploying in restricted environments. The MCP abstraction is well-suited for read-only monitoring and safe automation; write operations can be disabled via AF_READ_ONLY.

Install

astro-airflow-mcp on PyPI

pip install astro-airflow-mcp

uv add astro-airflow-mcp

poetry add astro-airflow-mcp

Installing astro-airflow-mcp

Before you install

Low friction install with a pure-Python wheel and eight runtime dependencies including fastmcp, pydantic, and httpx. Actively maintained with a release 37 days ago. Requires Python 3.10 or later.

Quickstart

pip install astro-airflow-mcp

# Configure for Claude Desktop
# Add to ~/Library/Application Support/Claude/claude_desktop_config.json:
{
  "mcpServers": {
    "airflow": {
      "command": "uvx",
      "args": ["astro-airflow-mcp", "--transport", "stdio"]
    }
  }
}

# Or run standalone with custom Airflow instance:
export AIRFLOW_API_URL=http://localhost:8080
export AIRFLOW_USERNAME=admin
export AIRFLOW_PASSWORD=admin
uvx astro-airflow-mcp --transport stdio

Requires Python 3.10 or later. Airflow instance must be running and accessible at the configured AIRFLOW_API_URL (defaults to http://localhost:8080).

Verify before relying

• Whether the package works with all Airflow 2.x and 3.x versions or only specific minor releases
• Performance characteristics when managing large numbers of DAGs or task instances
• Whether MCP resource and prompt features are fully documented and stable

Package facts

Evidence: astro_airflow_mcp-0.9.1-py3-none-any.whl

Tags

More Distributed Computing packages

Further reading