Introducing tldraw for ChatGPT and Codex

Today we're launching the tldraw plugin for ChatGPT and Codex.

It’s one of the first to use OpenAI’s new MCP spec. The plugin gives your chat a tldraw canvas, so you, your agents, and your teammates can all collaborate together in a real-time canvas.

Plus, you can tag your agents in comments, just like you would your teammates, to give them location-specific context.

Every file is available as multiplayer, so you can bring your whole team in.

The plugin connects to your free-to-use tldraw.com account. Every board you make in ChatGPT or Codex is a real tldraw.com file, and all your existing boards are there too.

To get started, go to ‘plugins’ in ChatGPT/Codex, and search for tldraw. Install the plugin and sign in with your tldraw.com account — or create one if you’re not already signed up. Then open a new chat and try a prompt like, ‘draw a diagram of how traffic lights work’ or ‘create a magazine-style flowchart of what I should read next’.

A canvas in your sidebar

When you first open tldraw, you can see all of your boards or switch between workspaces, as long as you’re logged in. If not, you’ll see a prompt to make a new file or some sample prompts. Click on a thumbnail to open a board or choose to create a new one.

Then ask your agent to fill it. Then, ask your agent to fill it with whatever you need - design iterations, workflows, charts, anything goes. We’ve shared some use cases below and will be sharing more over the next few days and weeks.

You can also bring boards into a conversation. Type @tldraw to search your boards and attach one as context. Your agent will read what's on it, whether that's an architecture diagram from last quarter, a wireframe or a page of sticky notes, and use it for the task at hand.

Agents and humans working together

Your agent isn't the only one on the board. Anyone you share a board with can join, and you'll see their cursors as they work. The plugin tells the model who else is present, so it knows when it's working alongside other people.

Agents can also use comments, the same way your teammates do. Leave a comment on part of the canvas and ask your agent to act on it. The agent can read comments and their attachments, reply in threads, and leave comments of its own.

This works across agents too. Two people on the same board, each with their own Codex, can both edit the canvas at once, and the changes sync to everyone.

How the tldraw ChatGPT plugin works

The plugin is built on OpenAI's new extension to the Model Context Protocol (MCP), which lets a plugin run an interactive interface in the sidebar, rather than a widget that has to reload each time.

On our side is a tldraw MCP server that gives the model the ability to:

• create, open and rename boards
• search through your existing boards
• look at the canvas, by taking screenshots and reading its shapes, text and comments
• draw and edit on the canvas, including Mermaid diagrams

Alongside the tools, the plugin sends the model context about what you're looking at, like which boards are open, what you've selected, and who else is on the board. That's why you can say "make this bigger" or "connect these two", and the agent knows what you mean.

Boards sync through the same system that powers multiplayer on tldraw.com. Anything your agent draws is saved to your account straight away, and edits from anywhere, whether ChatGPT, Codex or tldraw.com in a browser, show up everywhere. Use the Copy Link button to jump to the full editor at any time.

What you can do with the tldraw ChatGPT plugin

Some things are easier to draw than to describe. Here are just a few things we’ve tried with it:

Turn a wireframe into an app

Sketch a rough layout on the canvas, with boxes, arrows and a few labels. Then ask Codex to build it. When you want to change something, move things around on the canvas instead of writing a paragraph about it.

Annotate your app

Drop in a screenshot of the thing you're building. Circle the button that's in the wrong place, cross out the section you don't need, and scribble a note. Then ask Codex to update the code from your annotations.

Diagram your codebase

Ask Codex to map out a system, a data flow or a state machine. If the diagram is wrong, fix it on the canvas and ask the agent to update the code to match. It works in both directions.

Explain something visually

Ask ChatGPT to explain a concept as a diagram rather than a wall of text, then keep asking questions against the drawing.

We’d love your feedback. Get in touch with us on our Discord or follow us on X. If you'd like to put a canvas like this in your own product, check out our docs or starter kits.