AI Catchup

Cursor Agent SDK: Build, Evaluate, and Deploy Agents as Code

By 1 min read

Cursor's Agent SDK gives teams a code-first way to define, test, evaluate, and deploy AI agents. Its docs describe project instructions, tools, and evaluations as versionable files, with Markdown and TypeScript in the project. Developers can use Cursor-managed hosting or self-hosted infrastructure.

Cursor's Agent SDK gives developers a code-first way to define, test, evaluate, and deploy agents. Its docs describe Markdown and TypeScript files that combine deterministic code with model judgment, plus Cursor-managed or self-hosted deployment. (Cursor Agent SDK documentation)

Agents defined in files

The Agent SDK docs describe each agent as a project whose instructions, tools, and evals can be reviewed, tested, and versioned like code. Markdown and TypeScript combine deterministic code with model judgment, while the SDK's APIs connect agents to places such as Slack, pull requests, issue trackers, and Cursor. (Cursor Agent SDK documentation)

Cursor-hosted or self-hosted

Cursor documents two deployment choices: use Cursor-managed hosting, or run the agent on infrastructure you manage. (Cursor Agent SDK documentation)

The documented quick start installs the CLI, scaffolds a project, and starts a development session. Cursor's setup instructions call for Node.js 22.13 or newer. (Cursor Agent SDK documentation)

npm install --global @grok/agent-sdk
agent-sdk init ./my-bot
cd ./my-bot
agent-sdk dev

The setup guide also includes a browser sign-in step before initializing an agent project. (Cursor Agent SDK documentation)

Agent SDK and TypeScript SDK

Cursor's TypeScript SDK is a separate developer surface. Its documentation describes the @cursor/sdk package as a way for TypeScript to call the same Cursor agent available in the IDE, CLI, and web app. The Agent SDK documentation instead focuses on versioned agent projects made from instructions, tools, and evaluations. (Cursor TypeScript SDK documentation; Cursor Agent SDK documentation)

Sources

Related coverage

Frequently Asked Questions

What does Cursor's Agent SDK do?

It lets teams define, test, evaluate, and deploy agents as code, with instructions, tools, and evaluations stored as project files.

What languages and files does the Agent SDK use?

The docs describe Markdown and TypeScript files that combine deterministic code with model judgment in an agent project.

Where can Agent SDK projects run?

Cursor documents both Cursor-managed hosting and self-hosted deployment.

How is the Agent SDK different from Cursor's TypeScript SDK?

The Agent SDK docs describe versioned agent projects for building and deploying agents as code. The separate `@cursor/sdk` package lets TypeScript call the same Cursor agent available in the IDE, CLI, and web app.

Get an email when Cursor changes its plans, prices or limits

We check every change against Cursor's own pages before we send. You also get the weekly AI Catchup; one click unsubscribes.