Magpie
A desktop app that turns any website into clean data: scrape, crawl, extract, watch and research the web from your own machine, on your own API keys, for one flat price with no credits. The hard part was a scraper that holds up on real pages without renting someone else's cloud.
- Project
- Magpie
- Year
- 2026
- Stack
- Go, Wails, React, TypeScript
- Role
- Founder, Design & Build
The brief
Magpie is two codebases on purpose. The engine is an open-source Go library and CLI: one static binary, no CGO, that fetches a page, strips the boilerplate and returns clean markdown, with schema-driven extraction on top. The desktop app is a closed Wails shell over it, a React and TypeScript interface that ships for Windows, macOS and Linux.
The split keeps the hard part honest. Anything the app needs from the engine lands as a public change to the core first, so the scraper behind the product is the same one anyone can read, run and audit.
A real browser only when the page needs one
Every fetch starts static and fast. When a page is built by its scripts or answers with a bot challenge, Magpie hands it to a real Chromium on your machine, so the content arrives whole: clean markdown, JSON or a screenshot.
Describe the data, get typed JSON
Write a schema, or say what you want and let Magpie draft one fitted to the page, then get a typed record back through whichever LLM provider you bring. Every run shows its tokens and cost, and extraction is capped, so it never becomes a surprise bill.
Built for agents, too
The app serves an MCP endpoint on localhost, one click from Claude Code, Claude Desktop or Cursor, so an agent scrapes and extracts through the same engine you drive by hand.