# Doop > Source: > Published: 2026-11-01 00:55:32+00:00 Doop is an infinite canvas where AI agents design beside you. They join through MCP — as **yours**, via OAuth — sketching frames, filling real copy and reviewing their own work while you steer. Agents sync their tasks to the canvas as they work, so you can watch anyone's agent mid-design — not just your own. Frames stream in character by character, comments turn into tasks, and every decision stays visible to the next agent that joins. Love it. Can we try a morning version, darker and warmer? Agents can see this screen is approved for implementation. Approved by Ana · 14m ago Mark a frame ready for dev and an automation takes it from there: Doop drafts every ad size, the Copywriter localizes them, Brand Compliance checks the lot — on the same canvas, as drafts for you to sign off. Or put it on a schedule, and every Monday last week’s ads come back with fresh variants. Doop is an MCP server. Point Claude Code, Cursor, Codex — any agent that speaks MCP — at your canvas and it designs as you, on your account and your subscription. No lock-in, no middleman. The canvas goes where you go, runs when you're not there, and never keeps your work inside. Doop for iPhone. The same canvas, the same agents, the same live cursors — approve a frame from your pocket while the agent keeps streaming on your desk. Schedule agents to run without you. A feature image for every new blog post, fresh OG images each Monday, a visual review on every pull request — set it once, watch it land on the canvas. Doop is open source. Read the code, run it on your own infrastructure, or send a pull request — the canvas you design on is one you can own. Every frame is a file and a URL. PNG, PDF, HTML, Canva, Figma — or a live image link that re-renders when the design changes, so what you shared is never stale. Ask for a background, an illustration, a product render, a mascot — and Doop draws it straight into the frame, in your palette. GPT Image 2 draws today; the next generation of image models lands in the same canvas. Paste a URL, drop an HTML export, pull in a Figma file, a Canva design or a PSD, or connect a GitHub repo. It lands on the canvas as an editable snapshot — the original stays untouched, and agents design the next version right beside it. Connect a repo and Doop opens the pull request. The frame becomes real HTML and CSS in your codebase, in your conventions — hand it to Claude Code or Cursor to finish, or merge it as it is. No handoff doc, no redlines. Anything else — ask in Discord, or open an issue on GitHub. The whole thing is open source. Any agent that speaks MCP. Claude Code, ChatGPT, Codex, Cursor, Gemini, Kimi, Qwen, DeepSeek and more connect in one step: copy the MCP URL, approve once in the browser, and the agent gets its own cursor on your canvas. Doop is the closest thing. It gives ChatGPT, Codex, Gemini, Kimi and Qwen the same connect-an-agent-and-watch-it-design workflow that Claude Design gives Claude — on one shared canvas, with your own subscription, and open source. Yes. Like Paper, Doop is an infinite canvas where AI agents design real HTML. On top of that it is live multiplayer, works with any MCP agent instead of one, imports websites, Figma, Canva and PSD files, opens GitHub pull requests, and is open source you can self-host for free. Those tools generate screens from a prompt inside their own app. Doop is a shared canvas that your own agent works on through MCP: it edits the HTML directly, streams the result live, remembers your taste and opens pull requests — with whichever model you already pay for. Yes. Add `https://doop.design/mcp` to Claude Code, approve once in the browser, and Claude Code gets tools to create frames, stream HTML, screenshot its own work, import sites and read the canvas memory. Codex, Cursor and every other MCP client get the same tools. Doop is free while in beta and open source under the MIT license on GitHub. Self-host it as a single Docker container plus Postgres, or use doop.design. No. Your agent keeps running wherever it already runs, on whatever subscription you already pay for. Bring your own key only if you want image generation or the built-in Doop agent on your own account — nothing is billed through Doop. Connect a GitHub repo and Doop opens a pull request with the frame as real HTML and CSS, in your project's conventions. Or export any frame as PNG, PDF, HTML, Canva or Figma — or embed its live image URL, which re-renders when the design changes.