AI NewsOpen sourceAnnouncement
Doop is an open-source design canvas where agents draw beside people
An AGPL-licensed multiplayer design canvas that presents itself as an open alternative to Paper.design, where AI agents join over the Model Context Protocol and build frames while people watch, at 571 stars in 30 days.
Image: GitHub
Why it mattersDesign work with an agent usually happens in a chat window and arrives as a finished file, and this puts the agent on the canvas where you can interrupt it halfway through.
Doop is a multiplayer design canvas that people and AI agents use at the same time. The project describes itself as "the open-source alternative to Paper.design," a commercial design tool. It has 571 stars on GitHub over the past 30 days and is licensed under the GNU AGPL v3.
How agents take part
Agents connect over the Model Context Protocol. The server exposes an endpoint at /mcp over stateless HTTP with more than 16 tools behind it, covering canvas management, frame operations, and screenshot rendering. A frame is an artboard holding HTML, and an agent can create and edit them.
Two details make this different from asking a model for a design and getting a file back. The first is that agent output streams onto the canvas live, so a person watches the design being built rather than receiving it finished. The second is get_frame_screenshot, a tool that lets the agent render and look at its own work before continuing. Human collaborators can give feedback while a frame is still being built.
Authentication runs through OAuth, and the project says a connecting agent inherits the access of the human who authorised it, rather than holding separate permissions on the canvas.
Running it
Doop includes a built-in agent that uses Anthropic's Claude by default, with Azure OpenAI or a user's connected ChatGPT subscription as alternatives. The rest is standard self-hosting: an embedded PGlite database by default, external Postgres supported, and a Docker Compose path for production. The human side of the canvas has the collaboration features you would expect, with live cursors, presence, and an activity feed of every edit.
The project is early. The README shows 77 commits on the main branch and no version tags.
The interesting part is how corrections work. When a model produces a design in a chat window, the only point for feedback is at the end, and a wrong design means generating it all again. When the work appears on a canvas as it is made, someone can say "stop, the spacing is wrong" at the point the mistake happens. That is a different working relationship, and it is closer to how design review already works between people.
The licence deserves a look before anyone starts to depend on it. AGPL v3 is a strong copyleft licence, and if you run a modified version as a network service it obliges you to offer the source of your modifications to users of that service. For an internal design tool that is usually not a problem. For anything you plan to build a product on, read it properly first.
At 571 stars, 77 commits, and no releases, this is worth an evaluation rather than a migration. It is also exactly the kind of thing that rarely gets announced: an open replacement for a commercial tool has no vendor with a reason to tell you it exists.
Source
Primary source: Doop on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.