justdraw

An AI artist for vector art.

Connect Claude, ChatGPT or Cursor once, then just ask for a drawing. You get real, editable shapes — not a picture of one.

Connect AI

justdraw speaks MCP. Add this server once, and your AI can make drawings and edit them — real vector shapes you can then open and change yourself, not a picture.

1 · Add the server

Claude Opens the connector dialog with justdraw filled in. Confirm and you’re done. Cursor Installs it in one click. VS Code Installs it in one click.
Claude Code
ChatGPT Settings → Connectors → Advanced → Developer mode, then Create and paste the URL. Needs a paid plan.

2 · Ask for a drawing

Once it is connected, say this to your AI:

Draw me a sunny day in California in justdraw: a big yellow sun in a blue sky, three green palm trees on golden sand, and the blue ocean behind them. Then give me the link.

It will make the project, draw it, and hand back a link. Open the link and every shape is yours to move, recolour and export.

Questions

Can an AI agent actually draw here, or just generate a picture?

It draws. An agent sends the same JSON commands the editor’s own buttons send (element.draw, group.create, arrange.align), so what it produces is a real vector document: shapes with coordinates, fills and ids, sitting in layers, in your undo history. If a circle is four pixels too far left, you drag it four pixels right — because there is a circle. An image generator returns pixels, and there is nothing to drag.

How does an agent learn to use it?

Paste a project link at it. GET /p/<uuid>.md returns a briefing generated from the live project: what is currently on the canvas with real ids, the whole command surface, and curl examples with that project’s id already filled in. An agent that has never heard of justdraw.fyi can be handed a URL and get to work.

Do I need an account?

No. There is no login and nothing to install. You press one button and get a URL, and the drawing lives at that URL. Anyone holding the link can open and edit it — the UUID is the capability — and nobody who doesn’t have it can find the drawing.

Is there a command line?

Yes. jd drives any project from a shell, and every command prints the id of the thing it made, so ids pipe from one command into the next: jd draw <project> rect 0 0 40 30 prints an id, and jd style <id> fill=#5b7cfa colours it.

Why does a drawing command need the project open in a browser?

Because the browser is the engine. The editor already knows what “align these three shapes” means, so the server relays commands down to the open tab rather than reimplementing the geometry in Go. That means there is exactly one implementation of every command, shared by the toolbar, the keyboard, the CLI and any agent — they cannot drift apart. Read-only commands work with no tab open.

What can it actually do?

Bezier pen, curvature and smooth tools, boolean shape building, knife, scissors, reshape, variable stroke width, layers, groups, artboards on an infinite canvas, and export to SVG, PNG, PDF and YAML. It is a real vector editor, not a toy that happens to open fast.

There is a lot more in the blog — a post on every tool, and on why an agent's edit is the same edit as yours.