Cursor Canvas Guide: What It Is and What to Build With It

Cursor's June 2026 Canvas update added Design ModeA way to point at an element in Cursor's built-in browser and change it directly, instead of describing it in words. Press Enter for the full definition. inside canvases, an interactive context usage report, a Debug with Agent button, full-screen shared canvases and embedded canvas buttons that can run prompts.

Knowing what shipped is the easy part. Turning Cursor canvases into something a team relies on takes a few deliberate steps, mapped below.

1. 1Use a canvas when the output is an artifact, dashboard, audit or decision report.
2. 2Ask for context usage reports when the agent feels slow, expensive or distracted.
3. 3Use Design ModeA way to point at an element in Cursor's built-in browser and change it directly, instead of describing it in words. Press Enter for the full definition. to fix canvas UX without describing every element in prose.
4. 4Share the canvas only after labels, caveats and permissions match the audience.

New capability comes with new ways to get it wrong. Keep these boundaries in place as you adopt it.

Enriched UI, from a plain prompt. Open Canvas from the plus menu and Cursor renders a rich, pretty visualization using its built-in UI frameworks - you don't tell it to build in React or hand over brand guidelines. One field engineer calls it diagrams on steroids. The agent accepts an image, a CSV or your codebase as input and turns it into something you can look at.

• Journey maps - "use Canvas to walk through and display the different user journeys of my paycheck planner app" produced a four-core-journey journey matrix.
• Documentation - interactive views of how a codebase is built, plus dependency and dashboard health.
• Onboarding material - one FE fed Canvas the Cursor docs as context and generated a getting-started pamphlet for a new enterprise customer.
• One-off dashboards - a growth engineer turned an anonymized CSV usage export into a live dashboard, then iterated with follow-ups ("match the Cursor dashboard", "add a tooltip").

Four conditions have to hold, and a solo account fails two of them.

Publish lives on the canvas toolbar and uploads a live snapshot that teammates open in a browser, with the same layout, charts and tables. Shares are available on paid plans only, Pro, Teams and Enterprise; free accounts cannot create them. Because every share is team-visible you also have to be on a team, though a Pro user on a team can share. And sharing needs a privacy mode that allows data storage, so Legacy Privacy ModeCursor's setting that guarantees code data is not used for training by Cursor or its model providers, and that an admin can enforce org-wide; data-retention terms are a separate, contractual layer. Press Enter for the full definition. blocks it outright.

The last one is what catches enterprises rather than individuals, since it is a privacy decision rather than a sharing decision, and the person clicking Publish is rarely the person who made it. Team admins can also switch Shared Canvases off for the whole organization from team settings. Sensible control, and worth checking before you promise somebody a link.

Older write-ups on canvases, including the workshop notes further up this page, describe publishing to Notion and treat link sharing as coming soon. That gap has closed. Browse what your team has published from Shared Canvases on the dashboard.

Publishing again refreshes the existing share rather than minting a second one. Useful, mostly, since the link you pasted into a doc keeps working. The consequence is that a caveat you left off the first version stays off for everyone reading it until you republish, which is the strongest argument I know for putting the source note and the date into the canvas itself instead of into the message you sent alongside it.

Cursor's guidance here is blunter than most iteration advice. For anything larger than a tweak, revert and prompt again with more detail, because that is usually faster than nudging your way there through small follow-ups. Describe the change rather than editing the layout by hand. Manual source edits are for the small stuff.

There is a second move in the docs that people skip, and I would argue it is the one that matters most. If the numbers look stale or wrong, ask Cursor to rerun the underlying query or show its work. The tempting alternative is to correct the figure in the source, which fixes what the canvas displays and leaves the query that produced it just as wrong as it was, ready to be rerun later with fresh data by somebody who trusts it.

That is the part worth internalising about canvases generally. A canvas is not a picture of an answer; it is a saved thing you reopen and rerun, and it keeps whatever query it was built on.

The second time somebody asks for the same shape. A canvas skill carries four parts: a trigger description so Cursor knows when to reach for it, layout instructions naming the sections, stats and tables, the data sources or queries that populate it, and formatting rules such as units, date ranges and sort order. After that a short prompt regenerates the canvas with fresh data, and every teammate who invokes the skill gets the same output shape.

For your own work the skill is often overkill, because canvases are saved into the workspace's canvas list and you can reopen one and rerun it. The reason to package it is other people.

Formatting rules earn their place faster than the layout instructions do, which surprised me. Two versions of the same dependency audit, one sorted by severity and one covering a different date range, look like a disagreement about the codebase rather than a difference in prompts. Pinning the units and the window costs a line in the skill and removes the entire argument.

A skill will not fix the limit that actually matters, though. A canvas is an artifact rendered next to the chat, not a file in the repo, so if the output is meant to be reviewed, merged or depended on by other code, it belongs in a pull request. Shared canvases make good evidence for a decision. They do not replace the source or the review of it.