MotherDuck now allows AI agents to autonomously provision a data warehouse account via a single API call — no forms, credentials, or human intervention required. An agent sends a POST request to https://new.motherduck.com and receives a token, a claim URL, and a quick-start guide. The post walks through a full example using Claude Code: querying GitHub Archive star events, running sub-second analytics with DuckDB, and packaging recurring ingestion logic as a hosted Python runtime job (Flights). Humans can later claim the provisioned organization via a one-time URL. The dlt team has already integrated the signup API to connect pipelines to MotherDuck warehouses without exposing credentials to the agent.
Questions this post answers
How can an AI agent programmatically provision a MotherDuck data warehouse without human intervention?
An agent sends an empty POST request to https://new.motherduck.com — no headers, body, or auth required. The response returns a motherduck_token for read/write access, a one-time claim_org_url for a human to later take ownership, a plain-text quick-start guide, and the cloud region (aws-us-east-1 only). The agent can then connect immediately using the DuckDB CLI or Python client with the injected token. Teams building agent-driven data pipelines track MotherDuck and DuckDB developments on daily.dev.
What are MotherDuck Flights and how do they work for scheduled data ingestion?
Flights are Python programs that run on MotherDuck's hosted runtime, either on demand or on a cron schedule. They support virtually any pip-installable Python code and use the DuckDB Python client to connect via `duckdb.connect('md:')`. MotherDuck injects credentials at runtime, so the agent writes only the pipeline logic and never handles secrets directly. Scheduled Flights are available on the Business plan. Developers automating analytics pipelines with hosted runtimes follow MotherDuck updates on daily.dev.