Summary
Between a raw fetch and writing a scraper.
A plain HTTP tool hands the model a page of markup; a scraper needs selectors that break. Here the schema is the specification and the model fills it from the rendered page, so the same schema survives a redesign.
What it is
A .NET server that turns a website into structured data. You supply the URL and a JSON Schema describing what you want; it drives a browser through Playwright, works over what it finds with Azure OpenAI, and returns JSON matching your schema.
What you get
- A page extracted into the exact shape you asked for - you write the schema, it fills it
- Optional instructions alongside the schema, to steer how the page is worked through
- A real browser doing the navigating, so pages that only assemble themselves when rendered are not a special case
- The call declared read-only: it navigates and extracts, it does not act on the site
Requirements
The .NET server, an Azure OpenAI deployment for the reasoning step, and Playwright for the browsing. A cloud version exists - access is by the request form linked from the repository.
Setup effort
One command plus a key — docker build -t skrapr ., then supply credentials