Introducing AI Agent Tooling: Bringing Airflow Intelligence to Your Local Workflow

AI coding agents are good at writing code, but they struggle with writing correct code for systems they don’t understand.

That’s the insight driving a new pattern in developer tooling: companies are releasing specialized agent skills that teach AI coding agents the nuances of their technology stacks. Following the framework Vercel developed for this approach, we’re releasing Astronomer’s open source AI agent tooling to bring deep Airflow and data engineering expertise to any AI coding agent.

Whether you use Claude Code, Cursor, VS Code, or any of the 25+ compatible AI coding tools, you can now access specialized Airflow knowledge right where you already work.

Why We Built This

Over the past year, we’ve learned a lot about what makes AI assistance genuinely helpful for Airflow development through building Astro IDE. We’ve seen how effective AI becomes when it deeply understands Airflow patterns—knowing the difference between Airflow 2.x and 3.x patterns, understanding operator-specific failure modes, and guiding developers toward current best practices rather than outdated approaches. The high code acceptance rates we’ve seen with Astro IDE prove the value of this deep contextual understanding.

But we also heard something important from the community, not everyone wants to leave their preferred development environment. Some teams are deeply invested in VS Code. Others have standardized on Cursor or Claude Code. Local-first workflows matter, and forcing a tool switch creates unnecessary friction.

So we asked ourselves, what if we could bring that same Airflow intelligence to wherever developers already work?

That question led to our open source AI agent tooling for Airflow.

What’s in the Repo

Our agent tooling packages our Airflow and data engineering expertise as installable skills and an Airflow MCP (Model Context Protocol) server. This means any AI coding agent can gain deep knowledge of Airflow best practices, Dag patterns, testing strategies, debugging workflows, and data warehouse operations.

The project includes:

Agent Skills — Modular capabilities that AI agents can invoke based on what you’re asking:

• Setting up Airflow projects with proper structure
• Writing Dags following current best practices
• Testing and debugging workflows locally
• Analyzing data warehouse schemas and running SQL
• Tracing data lineage upstream and downstream
• Annotating task lineage and creating custom OpenLineage extractors
• Profiling tables for quality assessment
• Migrating Dags from Airflow 2.x to 3.x

Airflow MCP Server — Full REST API integration that gives agents direct access to:

• Dag management and triggering
• Task logs and execution details
• System health monitoring
• Auto-discovery of local Airflow projects
• An agent-first CLI, af, for interacting with Airflow

What makes this tooling effective is how it applies context. Skills adapt their recommendations based on your Airflow version and installed providers, guide you toward community-recommended patterns like dynamic task mapping, and pull real execution data when investigating issues rather than making assumptions. The agent doesn’t just generate code, it helps you iteratively test and validate changes in your local environment before deployment.

Beyond just Airflow skills, the data warehouse and analysis capabilities included here are informed by real production use. At Astronomer, we’ve built and productionalized an internal AI data analyst that our teams query via Slack or the CLI for everything from table profiling to business metrics. These agent skills reflect what we’ve learned works in practice. We’ll be sharing more about this approach soon.

One Command to Get Started

The entire setup takes one command:

npx skills add astronomer/agents --skill '*'

This installs Airflow and data engineering skills into your project via skills.sh, making them available to Claude Code, Cursor, and other AI coding tools. For MCP clients like Claude Desktop, installation is equally straightforward:

claude plugin marketplace add astronomer/agents
claude plugin install data@astronomer

Once installed, you can work naturally. Ask your AI agent questions like:

• “Create a Dag that loads data from S3 to Snowflake daily”
• “What tables contain customer data?”
• “Why did my ETL pipeline fail yesterday?”
• “Show me revenue trends by product category”

The agent automatically invokes the right skills to help you accomplish these tasks, using the same patterns and best practices we’ve refined through customer learnings with Astro IDE.

Airflow Intelligence, Wherever Your Work

We believe teams should be able to develop Airflow workflows in whatever environment works best for them, whether that be in-browser or in their local environment.

Astro IDE remains the best choice for teams who want a purpose-built, fully integrated browser-based experience in Astro. You get an Airflow-native AI assistant with zero-setup local development, in-browser testing, and one-click deployments to Astro, all optimized specifically for Airflow workflows. For many teams, especially those onboarding new data engineers or working across distributed teams, this integrated approach removes friction and accelerates productivity.

Our open source AI agent tooling serves teams who prefer local-first development. If you’re already deeply invested in VS Code, Cursor, or Claude Code and switching workflows feels disruptive, you shouldn’t have to give that up to get expert Airflow assistance. The agent tooling brings the same intelligence into your existing environment.

Some teams will standardize on one approach. Others will have different team members using different tools based on their experience level and preferences. The choice comes down to how you and your team work best. We’re not prescriptive about it, we just want to meet you where you are.

Open Source and Community-Driven

We’ve released our AI agent tooling as fully open source under the Apache 2.0 license, hosted on GitHub at github.com/astronomer/agents.

We want this to be shaped by the community’s needs. If there’s a workflow you’d like skills for, a pattern you think agents should understand better, or improvements to make installation easier, we want to hear about it. Open an issue, submit a PR, or start a discussion.

The data engineering ecosystem thrives when we build openly and collaboratively. This agent tooling is our contribution to making AI-assisted development accessible to everyone working with Airflow, regardless of which tools they choose.

Get Started Today

Ready to try it yourself? Here’s how:

1. Install the skills:

npx skills add astronomer/agents --skill '*'

2. Configure your data warehouse (optional but recommended): Create ~/.astro/agents/warehouse.yml with your connection details

3. Start building: Ask your AI agent Airflow questions and watch the skills activate automatically

You can check out the full documentation and examples in the GitHub repository.

We’re excited to see what you build!