Cursor has made its Agent SDK generally available, organizing prompts, typed tools, skills and MCP connections into a single project. Agents can run on pull requests, Slack messages, webhooks and scheduled events, while follow-ups resume the same conversation. The SDK includes a local playground, evaluations and replay of real pull-request events, with deployment to Cursor infrastructure or a self-hosted environment. Its changelog emphasizes checking which tools ran and with what inputs, not only the final text response.