Event date: 2026-10-20
Cursor’s October 20 update introduces a generally available Agent SDK. Developers can define an agent in a code folder, run it locally, evaluate its actions and deploy it to Cursor or their own infrastructure.
Projects separate instructions, typed tools, skills, MCP connections and communication channels. Supported triggers include source-control events, Slack messages, webhooks and schedules. Evaluation checks can inspect which tools ran and their inputs; follow-up messages resume the same conversation.
Why it matters
This makes an agent easier to maintain like other software. A team could start with issue triage, record representative cases and test behavior before granting it responsibility for repository changes.
Source: Cursor changelog