Cursor Canvases: Interactive Agent-Built Interfaces
Cursor introduced Canvases, a new response mode that allows agents to generate interactive dashboards and custom interfaces instead of plain text or markdown. Built on a React-based UI library with first-party components β tables, boxes, diagrams, and charts β Canvases serve as durable artifacts that live in the Agents Window alongside the terminal, browser, and source control. The feature is available in Cursor 3.1.
Sources & Mentions
4 external resources covering this update
Cursor Introduces Canvases: A New Visual Interface for Agent Output
Cursor shipped Canvases on April 15, 2026 as part of Cursor 3.1, fundamentally changing how agents communicate results back to developers. Rather than returning walls of text, markdown files, or raw code blocks, Cursor agents can now produce rich, interactive visual interfaces β complete with tables, diagrams, charts, and custom logic β that developers can explore, iterate on, and reopen at any time.
What Are Canvases?
A Canvas is an agent-created artifact rendered through a React-based UI library. Cursor provides a set of first-party components β including tables, boxes, diagrams, and charts β that agents draw on to represent information visually. These artifacts are persistent: in the Agents Window, Canvases live as durable items in the side panel alongside the terminal, browser, and source control tabs, so a developer can return to a Canvas later, refresh its data, or iterate on it with follow-up prompts.
The intent is to replace the cognitive overhead of parsing long markdown documents or scrolling through flat chat responses. Rather than presenting information as plain text, Cursor agents can now build tailored interfaces suited to the specific task at hand.
Practical Use Cases
Code Review
When reviewing pull requests, Canvases allow an agent to logically group changes, surface what matters most, and present a rich browsable interface. The agent can generate pseudocode explanations for complex algorithm changes, displayed inline within the Canvas layout.
Incident Response and Observability
Integrated with MCP servers for tools like Datadog, Databricks, and Sentry, agents can pull observability data from multiple sources and consolidate it into unified charts and dashboards within a single Canvas.
Research and Experiment Tracking
During optimization or evaluation tasks, an agent can create a Canvas that visualizes its own experimental progress in real time β showing which hypotheses it is currently testing, what results have been produced, and what patterns are emerging.
PR Management and Agent Orchestration
Canvases can be used to coordinate and manage other agents β presenting a structured view of in-flight background tasks, their statuses, and their outputs.
Skills and Reusable Canvas Templates
Cursor also introduced Canvas skills: pre-packaged templates that encode a specific canvas-generation technique. The Docs Canvas skill, for example, teaches an agent to generate an interactive architecture diagram of a repository on demand. Teams can author their own Canvas skills.
Access and Iteration Workflow
Canvases can be opened via direct links in Cursor's response cards, through the Command Palette, or through the Agents Window tab menu. The feature is available in Cursor 3.1 in both the Agents Window and the editor.