Cursor Releases Agent SDK for Code-Based AI Agent Development
Cursor has launched the Agent SDK, allowing developers to define and deploy AI agents using Markdown prompts and TypeScript tools.
On October 27, 2026, the Cursor team officially announced the release of the Agent SDK, empowering developers to define, build, and deploy custom AI agents entirely through code.
Image source: Cursor (@cursor_ai)
Architecture and Core Workflow
The Agent SDK introduces a clean, developer-centric approach to structuring AI agents by separating operational instructions from execution logic.
- Markdown Prompts: Agent instructions and system behaviors are authored in Markdown files, making them easy to read, version-control, and iterate on.
- TypeScript Tools: Tools and capabilities used by the agent to interact with external APIs or execute programmatic logic are written in TypeScript, ensuring full type safety and seamless integration.
- Flexible Deployment: Developers can choose to deploy their agents on Cursor's hosted infrastructure or run them entirely on their own self-hosted environments.
Built-in Evaluation System
Measuring performance shifts during agent iteration is a persistent challenge. The Cursor Agent SDK addresses this by integrating a built-in evaluation framework directly into the development cycle.
This enables developers to continuously measure agent outputs, track accuracy, and catch regressions immediately when prompts or underlying tools are modified.
Developer Impact
This release marks a significant step for Cursor as it extends beyond an AI-powered editor into an extensible platform for agent engineering. Teams looking to build, test, and host specialized coding or workflow agents using standard TypeScript tooling will find this SDK a robust addition to their stack. Detailed setup guides and API references are available in the official documentation.