Doop Review (2026): Pricing, Features & Honest Verdict

Doop is an open-source multiplayer design canvas where AI agents draw frames beside you over MCP. We found it fast, opinionated and unusually well built for something this young, but it is 12 days old and paid plans are coming. Best for: designers already running Claude Code or Codex. Price: Free in beta (free plan: yes). Rating: 7/10

What is Doop?

Doop is an infinite design canvas that lets AI coding agents draw on it with you, live. You connect Claude Code, Codex or any MCP client to a single endpoint, and the agent starts streaming frames onto the same board you are working on. You watch sections appear as they are written. No progress spinner, no “generating your design” modal, just rectangles and copy landing in front of you while you drag things around.

Kevin Goedecke pushed the first commit on 22 August 2026 and the repository describes itself, bluntly, as “the open-source alternative to Paper.design”. Twelve days later it had 586 GitHub stars and 74 forks. That is a lot of attention for a project that shipped its v0.2.0 tag on the same day we looked at it. The pitch lands because the problem is real: most people using AI agents for design work are stuck copying screenshots between a chat window and a design tool, and losing every decision in between.

What Are Doop’s Key Features?

Frames that stream onto the canvas as agents write them

This is the part that makes Doop feel different. When an agent builds a hero section, the section renders the moment each piece lands rather than after the whole response completes. You see the layout resolving in real time and you can interrupt it. Compared to the copy-paste loop most people run today, watching a frame assemble itself is a meaningfully better feedback signal, because you can tell in four seconds that the agent has misread your brief.

Six specialist agents with actual job descriptions

Doop ships a roster rather than one generic assistant: The Generalist handles layouts and frames, UX Lead handles flow and hierarchy, Copywriter handles headlines and microcopy, Brand Compliance handles palette and type, Accessibility handles contrast and semantics, and Visual Polish handles spacing and alignment. Splitting the work this way is smart. A single prompt asking one model to nail layout, copy, brand and contrast at once produces mush, and everyone who has tried it knows that.

Comments that become the backlog

Leave a note anywhere on the canvas and the right agent picks it up, does the work and replies with a screenshot. Your feedback is the task queue. There is also a built-in headless renderer that hands agents screenshots of their own frames so they can review fit, spacing and contrast before you ever see the result. Self-review by screenshot is a good idea, and it is the sort of thing that only works if the canvas and the renderer are the same system.

Self-hosting on one Docker container

Doop runs as a single Docker container plus Postgres. That is a low bar for self-hosting, and it means an agency with a client-confidentiality problem can run the whole thing internally. The terms are explicit that self-hosted deployments are governed by the source code’s licence rather than by Doop’s terms of service, which matters more than it sounds. More on that below.

How Much Does Doop Cost?

Doop is free. There is no paid tier, no credit pack and no seat price, because the product is in beta and the company has not started charging. You also bring your own model: Doop connects to whatever Claude or Codex subscription you already pay for, and takes no cut and adds no markup on top of it.

Compare that to Paper.design, the product Doop names itself against. Paper’s free plan caps you at 100 MCP tool calls per week, which is roughly one serious afternoon. Paper Pro is $20 per editor per month, or $16 per editor per month billed yearly, and lifts that to 1 million MCP calls a week plus video export and 100 MB images. So the honest comparison today is $0 against $20 per editor, with Doop carrying the risk that the price is not zero forever.

And the terms say so plainly. Doop’s terms of service, last updated 19 August 2026, state that the service is “currently in beta and free to use” and that the company will “give reasonable notice before introducing paid plans or materially reducing what free accounts can do”. That is more honest than most beta pages manage. It is still a warning. Nobody has published a number, so nobody can tell you what Doop costs in six months.

What Does the AGPL Licence Actually Mean for You?

Doop is licensed AGPL-3.0, not MIT. Most people read “open source” and stop there, and for personal use that is fine. But AGPL is the strongest copyleft licence in common use, and its defining clause treats network access as distribution. If you self-host Doop and let people outside your organisation use it over a network, you are on the hook to offer them the corresponding source, including your modifications.

For a solo designer or an internal team, this is a non-issue. For an agency planning to put a modified Doop in front of clients as part of a paid service, it is a real constraint that needs a lawyer rather than a blog post. The licence choice is also a business signal: AGPL is what you pick when you intend to sell a commercial licence later. Read it alongside the “paid plans are coming” line in the terms and the direction is clear enough.

How New Is Doop, Really?

Twelve days old, and the numbers show it. The repository has 586 stars, but the desktop releases tell a quieter story: desktop-v0.1.2 recorded 89 downloads, v0.1.3 recorded 9, and v0.1.4 recorded 2. That is about 100 desktop installs across every release against 586 stars. Stars are free to give and installs are not, and a roughly six-to-one ratio means most of the audience is watching rather than using.

The commit history is similarly thin on people. Kevin Goedecke has 85 commits; the next contributor has 2, and four others have 1 each. The GitHub organisation endpoint for the owner returns 404, so this is a personal account, not a company. There are 11 open issues, the code is TypeScript, and the last push was the same day as the launch. This is one person’s project with a good idea and early momentum, not a funded team with a support rota.

One more detail from the terms: total liability is capped at the greater of what you paid in the past 12 months or EUR 50. Since the product is free, that cap is EUR 50. Fine for a beta, worth knowing before you put client work on it.

Who is Doop Best For?

Use Doop if you already run Claude Code or Codex daily and you are tired of shuttling screenshots between a chat window and Figma. The setup is one command and one browser approval, and if you are already living in MCP the canvas slots into a workflow you have. It is also a strong fit for anyone who wants an agent-native canvas they can self-host on a single container without a procurement conversation.

Use Doop if the AGPL is not a problem for you, which for most individuals and internal teams it will not be.

Skip Doop if you need a design tool that will exist in the same shape a year from now, with a price you can budget and a support contract you can point at. Skip it if you are an agency planning to resell a modified version, because the licence makes that expensive. And skip it if you do not already pay for an agent subscription, because Doop supplies the canvas and not the intelligence. Without a model account attached, you are looking at an empty board.

Best Doop Alternatives

Paper.design is the direct comparison and the one Doop picked. It is the connected canvas for teams shipping with agents, it is more finished, and it has a company behind it. Free covers unlimited viewers and editors with 100 MCP tool calls a week and 25 MB images. Pro is $20 per editor per month, or $16 billed yearly, and raises the ceiling to 1 million MCP calls a week. If you want the same idea with an invoice attached, this is it.

Figma remains the default and is not really trying to be an agent canvas. The Starter plan is free with unlimited drafts and 150 AI credits a day up to 500 a month. Professional is $16 per month for a full seat, $12 for a dev seat and $3 for a collab seat. You get the entire ecosystem, the plugins and the handoff tooling, and you give up the live agent collaboration that is Doop’s whole reason to exist.

Excalidraw is the free-forever option if what you actually need is a fast shared canvas rather than agent choreography. The core editor is free and open source, and Excalidraw+ adds the hosted collaboration layer with a 14-day trial. It has none of Doop’s MCP machinery, but it will still be here in three years, which Doop cannot yet promise.

Final Verdict: Is Doop Worth It?

Doop is the most interesting thing we have seen in agent-assisted design this month, and it is free, so the cost of trying it is an afternoon. The streaming canvas is not a gimmick. Watching an agent build a layout in front of you changes how quickly you can correct it, and the six specialist agents are a better structure than one model doing everything badly at once. The Docker self-host path is a real advantage over anything venture-funded.

The reservations are all about age rather than quality. Twelve days old, roughly 100 desktop installs, one meaningful contributor, no named legal entity, an EUR 50 liability cap and paid plans that the terms openly say are coming. That is fine for experimenting and wrong for client deadlines. Install it, connect an agent you already pay for, and see whether the live canvas earns a place in your week. Just do not migrate a studio onto it yet, and read the AGPL before you put a modified copy in front of anyone who is paying you.