Doop

Doop is an infinite canvas where AI agents design alongside you. Instead of asking for a prompt and getting a one-off image, you connect an agent such as Claude Code or Codex through the Model Context Protocol (MCP), approve it in your browser, and it starts building real website frames on the canvas. The agent works as you, with attribution, and you can watch the design arrive section by section. There are no API keys to hand over and no extra AI subscription required.

The canvas is built for iteration. Frames render live as agents write them, so there are no loading spinners. You can leave a comment on any part of the design and the right agent picks it up, makes the change, and replies with a screenshot. Agents also review their own work: a headless renderer gives them screenshots of their frames so they can check spacing, contrast, and fit before you see them. Doop keeps a shared memory of decisions and taste — if you ask for softer corners or a single accent color, that preference is saved and applied to every new frame, and any agent that joins later reads it. You can also paste reference screenshots, and an agent will distill their palette, type, and mood into a design brief.

Doop is useful for designers who want to keep human control while delegating production work, for developers who already use Claude Code or Codex and want a visual output instead of code-only workflows, and for product teams that need quick landing page or marketing page iterations. Every frame gets a live URL, so you can embed it in a doc or tweet and it re-renders whenever the design changes. It's free in beta.