Cursor's Agent SDK is a framework for writing AI agents as code, testing them and deploying them. Cursor builds its own agents, such as Bugbot, with it.
In plain words
Each agent is a project folder: Markdown files hold its instructions and TypeScript functions are its tools, so changes get normal code review and version control. You provide the tools and instructions; Cursor runs the loop in which the model calls them. Agents react to events, such as a new pull request, can pass work to specialist agents or split it among subagents, and come with built-in evals (repeatable tests). One command deploys an agent to Cursor's servers, or you can run it yourself.
A quick example
Imagine a team that wants every pull request checked against its own coding rules. It writes an agent that wakes on each new pull request, reviews the change and posts a note on anything that breaks a rule.
Where you'll meet it
- Automated pull request review.
- Routine team work such as compliance checks and incident investigation.
Related
Sources
- DocsIntroducing the Agent SDK, Cursor · read 28 Oct 2026
- DocsAgent SDK, Cursor Docs · read 28 Oct 2026
Checked by WhatIsAI against the sources above · 28 Oct 2026. Spotted an error? Report it.