Cursor launches TypeScript SDK for coding agents with cloud sandboxes and token-based billing

On April 29, 2026, Cursor launched a public beta of its TypeScript SDK, which takes the company's coding agents out of the editor and into code and automation. The same agents that power Cursor's desktop app, CLI, and web version can now be spun up with a few lines of TypeScript — locally, in the cloud, or on your own workers.

One SDK everywhere

The new `@cursor/sdk` package gives developers programmatic access to the same runtime, harness, and models that Cursor's agents are built on. In effect, the company is turning its editor AI stack into an infrastructure layer: you can create an agent from a Node.js app, a script, or a pipeline and then talk to it like a service. Cursor points out that the SDK is already available to all users in public beta, and getting started comes down to a regular `npm install` and a call to `Agent.create()`.

Local execution isn't the only focus. The SDK's cloud sessions run on the same platform as Cursor Cloud Agents: each agent gets its own virtual machine with a strict sandbox, a clone of the repository, and a ready-to-use development environment. Such an agent can keep working on a task even if your laptop goes to sleep or the connection drops. Developers can stream progress, reconnect to the session later, and, once the work is done, get a finished branch, pull request, demo, or screenshots.

Tools inside the agent

The release's strength is that Cursor isn't offering a "lite" API version — it exposes nearly its entire agent loop. The SDK inherits codebase indexing, semantic search, and fast grep; it can connect MCP servers over stdio or HTTP, automatically picks up skills from the repository, and supports hooks for controlling the agent's work loop.

Subagents get special attention: the main agent can delegate subtasks to specialized helpers with their own prompts and models. Developers can also route tasks to any model Cursor supports, including its in-house coding model Composer 2.

• Dedicated sandboxed VMs for cloud sessions
• MCP connections via `.cursor/mcp.json` or inline config
• Hooks via `.cursor/hooks.json` for observing and extending the loop
• Subagents with separate roles, prompts, and models
• Automatic branch and PR creation once a task is done

Cursor also puts emphasis on usage economics. In the announcement, the company says the SDK is billed under the standard token-based consumption model. Its pricing page clarifies that plans already include a certain amount of model usage, and anything beyond that is billed on demand based on actual consumption.

For teams, that's an important signal: the SDK isn't sold as a separate enterprise product with a new license, but as a way to bring existing agent infrastructure into CI/CD, internal services, and customer-facing apps without rebuilding the whole system from scratch.

Scenarios for teams

Cursor pitches the release not as a demo toy but as a tool for production use. In the company's examples, programmatic agents run directly from CI/CD to summarize changes, find the root cause of CI failures, and update pull requests with ready-made fixes.

Another class of use cases is internal platforms, where non-engineering teams can query product data without writing code. An even more ambitious scenario is embedding the agent experience directly into customer products, so end users never have to leave the main app's interface.

"It's a path to our own agents in the same cloud runtime without managing VMs," is how

Faire described the idea behind the release.

To lower the barrier to entry, Cursor published a public cookbook repository with ready-made templates. It includes a minimal Quickstart for a local agent, a prototyping web app for launching cloud agents, a kanban board where you can hand a card to an agent and get a PR back, and a lightweight CLI for starting agents from the terminal.

This set shows the company is targeting not just the editor but the tooling layer around development — from interfaces for managers to automated work queues.

What it means

Cursor is taking another step from AI editor toward a platform for autonomous development. Coding agents used to live mostly inside the IDE; now companies can embed them into their processes like any other software component.

For the market, this strengthens the "agents as infrastructure" trend: the winners will be the players that offer not just a smart model, but also a secure runtime, control, integrations, and clear economics for running in production.