Cursor SDK Lets Devs Build & Run Programmatic AI Agents Locally or in Cloud

TL;DR

• Cursor has released its new Cursor SDK for building programmatic AI agents.
• Agents leverage Cursor's core runtime and models, runnable locally or on dedicated cloud VMs.
• Developers can get started with a few lines of TypeScript and npm install @cursor/sdk.
• The SDK is now in public beta and uses standard token-based consumption pricing.

Cursor has officially launched its new Cursor SDK, offering developers a powerful toolkit to build and deploy programmatic AI agents. This SDK grants access to the very same robust runtime, harness, and cutting-edge models that power Cursor's existing desktop, CLI, and web applications, bringing advanced AI capabilities directly into developers' workflows. The SDK is designed for flexibility, allowing agents to be crafted with just a few lines of TypeScript code.

A key advantage of the Cursor SDK is its versatile deployment options. Developers have the freedom to run their agents either on their local machines or securely within Cursor's cloud infrastructure. When opting for the cloud, agents operate against a dedicated VM and can leverage any supported frontier model, ensuring high performance and access to the latest AI advancements. This flexibility caters to various development needs, from rapid local iteration to scalable cloud-based operations.

Getting started with the Cursor SDK is straightforward. Developers can integrate the SDK by running npm install @cursor/sdk and utilizing the Cursor SDK plugin within their Cursor environment. To further assist with development, Cursor has made several sample projects available in a public repository. These examples serve as excellent starting points for users to fork, extend, and adapt for their specific use cases, accelerating agent development. Further details can be found in the TypeScript SDK Blog Announcement.

In conjunction with the SDK launch, Cursor has also rolled out significant updates to its Cloud Agents API. These enhancements include a rework for durable agents and per-prompt runs, improving how follow-ups, status, streaming, and cancellation are managed. The API now features first-class run streaming with SSE events and reconnect support via Last-Event-ID, alongside explicit agent lifecycle controls such as archive, unarchive, and permanent delete. The SDK is currently available to all users in public beta and is billed based on standard, token-based consumption pricing. For comprehensive documentation and to begin building, explore the TypeScript SDK Documentation.

Summary

• The Cursor SDK enables developers to build programmatic AI agents using Cursor's core runtime and models.
• Agents can be written in TypeScript and executed locally or on Cursor's cloud with dedicated VMs and frontier models.
• Easy setup via npm install @cursor/sdk, with sample projects available for rapid development.
• The SDK is in public beta, billed on a token-based consumption model, and comes with an updated Cloud Agents API for improved agent management.

Source: Build programmatic agents with the Cursor SDK · Cursor