ReviseMy

The itch

Agents are great at shipping and famously bad at knowing whether the thing looks right. Feedback usually means screenshots in chat, vague notes, or a sync—none of which give the agent a structured packet to act on. I’d had the idea since 2024 for a self-serve loop tool; it sat as an unused domain while I kept not building it.

The weekend

What finally unstuck it was a challenge: build on Laravel Cloud for the first time, in a weekend. Two days later the domain meant something—an MCP server your agent talks to from ChatGPT, Claude, or Cursor, with a review UI that makes human marks authoritative. Sometimes the missing ingredient isn’t the idea, it’s the deadline.

How the loop works

The product speaks in marks, not pins: yellow rectangles and M1/M2 badges on the capture. Drag a region or click a point. Guest feedback shows up in gray so you can tell it apart from yours. Optional second-opinion hints can queue a type-aware checklist, but they stay hints until you accept them—your marks are the ones that count.

Need another human without another account? Share a guest link you can expire (or rotate). Guests mark from the link; you stay the owner.

Every mark becomes a Board card: Open (you) → In progress (agent) → Resolved → Verified (you again). Agents attach before/after evidence when they resolve; request changes and open pass 2+ without losing context.

Try it

No account needed: grab a try token on the homepage, paste the MCP config into ChatGPT, Claude, or Cursor, and run a full agent↔human checkup in about two minutes. Open source, so the loop is inspectable too.

Tools & process