Agents that take apps apart
Paste a link or describe it. Our agents take it apart with REA, write a spec of how it behaves, and rebuild it from scratch: a real repo and a live preview. Or turn its public API into a Plugboard skill your AI agent can use.
Two ways out
A working version you own, built clean-room from how the app behaves: screens, flows, what it stores, what it calls. You get a repo you can run and deploy, a live preview, tests, and a report on what we rebuilt and what we left out.
A ready MCP server and agent skill, so Claude or any MCP-speaking agent can use that app. Built from the service's public API (or your own account through it), with setup docs, read-only by default, keys kept on your machine.
How it works
A sandboxed browser opens the public page like a visitor: records requests and storage key names, tries the main controls. No logins, no consent clicks.
REA attaches to the live page and reads its scripts: endpoints, routes, frameworks, storage.
An analyst writes what the app does: screens, flows, data. Never code, copy, images or brand.
Files, features and the builder: an AI agent when one is connected, otherwise a tested template.
The builder sees only the spec and writes new code with a new name, or generates the MCP server and skill.
Unit tests, a headless browser replaying the observed flow, or a real MCP client calling a tool.
A live preview on its own sandboxed origin, the repo zip and the report.
Clean-room by design
Two separate steps with a wall between them. The analyst looks at the app from the outside and writes a spec. The builder gets only the spec, without even the original's address, and writes fresh code. Brand words of the original are blocked and checked in the output.
Method and rulesRecently made
