Unbrowse swaps browser automation for APIs

Unbrowse is making a strong bet that agent browsing should look more like API discovery than screen control, and that’s a much more scalable idea than endless DOM hacking if the execution model holds up in the wild.

• –It reads cookies from local Chrome and Firefox sessions, which lets agents reuse real authenticated browser state without shipping credentials to a cloud proxy
• –Discovered endpoints get turned into reusable skills and published to a shared marketplace, so one agent’s reverse engineering can benefit the next one
• –The project is explicit about cost and latency: structured JSON replaces giant HTML payloads and screenshot-heavy browser loops
• –It still keeps a DOM fallback for sites with no usable API surface, which matters because not every workflow can be reduced to clean network calls
• –The open-source AGPL release and integrations with tools like Claude Code, Cursor, OpenClaw, and Windsurf give it credibility as real agent infrastructure rather than a one-off demo