The whole interface is a single call taking two official Airbyte connector names, which keeps the failure mode obvious: get the `source-` or `destination-` prefix right and you get working code, get it wrong and you get the wrong connector. Asking for `dataframe` as the destination is the fastest route from a source system to something you can actually poke at in Pandas.
A hosted MCP server that generates PyAirbyte pipeline code. You name an Airbyte source connector and a destination connector; it writes the Python, the setup steps and the credential scaffolding, using connector documentation and OpenAI for context. Anything in the Airbyte Connector Registry is fair game.
- Complete PyAirbyte pipeline code from a source connector name and a destination connector name — `generate_pyairbyte_pipeline`
- Setup and installation instructions alongside the script, plus environment-variable templates for the credentials the connectors need
- `dataframe` as a destination when the endpoint is Pandas analysis rather than a warehouse — the generated code hands you frames instead of a load step
Your own OpenAI API key, passed as `OPENAI_API_KEY` in the MCP server block. Nothing to install locally: the hosted server is at https://pyairbyte-mcp-7b7b8566f2ce.herokuapp.com/mcp. Cursor is the client the README targets, because at the time of writing it was the one passing env values through to a remote server; the config goes in `.cursor/mcp.json` for one project or `~/.cursor/mcp.json` globally.
One command plus a key — npx -y mcp-remote https://pyairbyte-mcp-7b7b8566f2ce.herokuapp.com/mcp, then supply credentials