Anysphere · 2026-06-04 · notable
Cursor 3.7 — Canvas Design Mode Lets You Select-and-Annotate UI Elements in Agent-Generated Canvases, Plus an Interactive Context Usage Report That Shows Where Every Token Went
Cursor 3.7 ships Design Mode — point-and-annotate edits to agent-generated UI canvases — and a Context Usage Report canvas that breaks down token spend across system prompts, tools, rules, and skills.
Cursor 3.7 lets you point at the wrong button in an agent-built UI canvas instead of describing the change in prose, and shows you where every context token went.
What is it?
A point release of the Cursor IDE shipped on June 4, 2026. The headline change is Design Mode for canvases — the interactive artifacts Cursor agents produce — letting users click and annotate UI elements directly so the agent can iterate on them. The release also adds a Context Usage Report rendered as an interactive canvas, full-screen browser viewing for shared canvases, and clickable prompt-trigger buttons agents can embed inside their own canvases.
How does it work?
Design Mode treats a canvas as an editable surface: the user selects an element (a button, a chart axis, a paragraph), annotates it inline, and the agent picks up the annotation as structured context for its next edit instead of having to parse a textual request. The Context Usage Report renders a context explorer canvas that breaks the current agent's token spend across system prompts, attached tools, project rules, and skills — and accepts follow-up prompts to drill in. Shared canvases now have a full-screen route so non-Cursor viewers can interact with the artifact in a browser, and agents can write canvases that include buttons mapping clicks to specific prompts.
Why does it matter?
Most agent tooling debugging is still 'tell the model what you mean in English about what you can see on screen.' Design Mode collapses that into a click. The Context Usage Report gives developers their first first-class view into where their context window actually goes — a recurring complaint as agents bloat their tool/skill payloads — and is useful for tuning rules and skills sets to bring cost down.
Who is it for?
Cursor users building UIs with agent canvases and teams trying to control context-window spend
Try it
Update Cursor → Settings → check for v3.7