← sources

I stopped describing UI bugs to my AI. I just point at them now.

source: reddit (reddit:ClaudeAI)published 2026-09-21 16:28ingested 2026-09-22 02:36

retained text

Every time I found a UI bug, I'd open Claude and type a paragraph: "the pricing page, the middle card, the button label is cut off ... no, not that button…" Half the time it fixed the wrong thing. And if I spotted five things in one pass, I'd forget four of them before I sat down to fix anything.

So I built Peekr. Press ⌥⇧2, drag over the bug or a suggestion, type one line. It lands on a list with the screenshot, the app and the URL. Later you tell your agent — Claude Code, Codex, Cursor, anything that speaks MCP — "fix what's open in Peekr," and it works through the list, looking at each screenshot.

The first question everyone asks is what happens to the screenshots, since they're literally pictures of your screen. Straight answer:

\- A capture is visible only to you until you file it into a group. Not even workspace admins can open it before that — the database enforces it, not the UI.

\- You can blur anything before sending, and it's burned into the image. The agent never gets the original.

\- Encrypted in transit. Never sold, never used to train models — mine or anyone else's. No ads, no third-party trackers.

Completely free, on Mac, web & Windows (soon). 

Honest question: what would stop you from using something like this?

Link's in the comments.

document fetched 2026-09-22 00:21 · canonical https://reddit.com/r/ClaudeAI/comments/1wmhozf/i_stopped_describing_ui_bugs_to_my_ai_i_just