Doop

Doop is an infinite design canvas where AI agents and people work on the same frames. Agents connect through MCP and OAuth, so they act as you, with your identity attached. It's built for designers and developers who already use Codex, Claude Code or another MCP client and want a place for those agents to design.

Your agent keeps running on the model you already pay for. Doop doesn't ask for API keys or sell another AI subscription.

• Streaming frames: sections render the moment an agent writes them.
• Comments to tasks: leave a note on the canvas and the right agent picks it up, then replies with a screenshot.
• Self-review: a headless renderer gives agents screenshots of their own work, so they check fit, spacing and contrast and fix problems before you see them.
• Live exports: every frame has a URL that re-renders when the design changes, ready for docs or og:image tags.
• Shared memory: tasks, decisions and comments stay on the canvas, so any agent that joins continues where others left off.
• Taste profile: your feedback is distilled into rules like corner radius, palette and shadows, applied to every new frame.

You can paste a public URL to import a site as an editable snapshot, then have agents design variants beside the original. Built-in image generation draws backgrounds and illustrations into frames using GPT Image 2.

Six specialist agents cover generalist design, UX, copy, brand compliance, accessibility and visual polish. Queue a card through the crew or @mention one in a comment.

Doop can be self-hosted as a single Docker container with Postgres.