Hi Product Hunt 👋 I'm a freelance designer from Kagoshima, Japan. I don't normally build backends — I build interfaces. For years I received change requests as spreadsheets. A screenshot pasted into a cell, a note beside it, and no way to tell which page or which element it was about. Working out where took longer than making the change itself. Then AI agents made prototypes fast. Suddenly we were reviewing working products instead of static mockups — but review itself hadn't changed. Still screenshots. Still "the button on the right, kind of near the top." So I built NeedBack. Reviewers open a URL, type a name, and click anywhere on your live site to leave a comment. No account, no extension. Each comment is stored with the URL, viewport, scroll position, the target element with its role and accessible name, the nearest heading, and a screenshot with the target outlined. The part I care about most: all of that is exposed over MCP. Point Claude Code, Cursor or Codex at your repo and say "check the open comments and fix them." The agent gets the context a human would have had — not just the comment text. Screen recording is in there too (90 seconds, with your voice over it). Since agents can't watch video, we pull timed frames out of the recording so they can still see what moved. Honest limits: desktop browsers only for now, and billing is JPY-only. There's a free plan, and the demo needs no signup at all. What I'd most like feedback on: is the context we hand to agents the right context? That's the part I'm least sure about.

NeedBack

Point at anything on the page. Your AI agent fixes it.

Point at anything on the page. Your AI agent fixes it.