Cursor now lets you use the same toolkit that it claims was used to build its agents, such as Bugbot and Security Review. If your team repeats the same PR review or bug triage every week, you can turn this into an agent stored in your repository alongside the rest of your code. With the Agent SDK, the agent becomes a folder. Instructions are written in Markdown, and the tools it can use are TypeScript functions—so everything goes through version control and code review like any other code. It comes with built-in tests. You can verify which tools the agent called, what data it used, and in what format it responded, then run these tests on your local machine or in your CI pipeline. The agent can be triggered by a PR, comment, or CI event on GitHub, GitLab, and Bitbucket; via mention in Slack; through a webhook; or on a scheduled timer. You can host it directly on Cursor or run it on your own server as a single Node.js process. Even when running on your own server, you still need a Cursor account to access its models.