Cursor's Agent SDK Is Now Generally Available: How to Build, Test and Deploy an Agent
Cursor's Agent SDK became generally available on October 20, 2026. It lets you define an agent as a folder of code, an instructions.md prompt plus typed tools, test it locally with evals, and deploy it to Cursor's own infrastructure or self-host it, per Cursor's changelog and documentation. [1]
SeekVero editorial · Published · Prepared with AI assistance; no hands-on testing claimed.
Prepared from cited sources. No hands-on testing claimed.
What Is Cursor's Agent SDK?
Cursor's Agent SDK is available starting October 20, 2026. It lets you define an agent as a folder of code, test it with evals, and deploy it to Cursor's own infrastructure or your own infrastructure. [1]
An agent is a project: instructions.md holds the prompt, each file in tools/ is one typed action the model can call, skills/ holds procedures the model loads on demand, mcp-connections/ adds tools from MCP servers, and channels/ defines where the bot receives and answers messages. [1]
How Do You Create and Run an Agent Locally?
Cursor's changelog describes creating a project with npx @cursor/agent-sdk init ./my-bot, then running agent-sdk dev to start the bot locally with a playground; passing --template starts from a security reviewer, an owners-policy approver, a PR autofixer, or a Linear or Jira triager. [1]
Cursor's own quickstart guide for building a PR reviewer instead installs the CLI with npm install -g @grok/agent-sdk and signs in with agent-sdk login before running agent-sdk init and agent-sdk dev. Both pages are official Cursor documentation, and the package name differs between them. [2]
That quickstart guide also states you need Node 22.13 or newer and a Cursor login for model turns. [2]
How Does an Agent Wake Up and Respond?
Agents wake on pull request, comment, and CI events from GitHub, GitLab, and Bitbucket; on mentions and DMs in Slack; on webhooks from any provider; and on a cron schedule, per Cursor's changelog. [1]
Each conversation keeps its own session and a complete record of every message and tool call; follow-ups on a pull request or Slack thread resume the same session, per the same changelog entry. [1]
How Do You Test an Agent Before Deploying It?
An eval sends a fixed message to the bot and asserts on what it did: which tools ran, with what input, and the shape of the reply. Cursor says to run agent-sdk eval locally or in CI. [1]
agent-sdk github replay turns a real pull request into signed webhook payloads and sends them to the bot, so GitHub-driven bots can be tested without waiting for a push, per Cursor's changelog. [1]
How Do You Deploy an Agent SDK Project?
Cursor's deployment docs describe committing and pushing the ref you want to run, then running agent-sdk deploy. The command builds the pushed ref, so uncommitted changes don't ship, and it waits until the agent is running before printing its URL. [1] [3]
The deployment docs say to declare the hosts and secret names your code calls, and to commit secret names but never commit secret values. [3]
For self-hosting, Cursor's docs say to run the Agent SDK under Node 22.13 or newer, and never under Bun. [3]
| Option or question | What to know |
|---|---|
| Privacy Mode | Available: Cursor stores deployment data the same way it does without Privacy Mode, per the deployment docs. [3] |
| Privacy Mode (Legacy) | Not available: deploy, a hosted session, and an eval run all refuse, per the deployment docs. [3] |
Frequently asked questions
Is Cursor's Agent SDK generally available?
Yes. Cursor's changelog says the Agent SDK is available starting October 20, 2026, and that it is generally available today. [1]
What do you need installed to build an agent with the SDK?
Cursor's quickstart guide for the Agent SDK says you need Node 22.13 or newer and a Cursor login for model turns. [2]
Does agent-sdk deploy ship uncommitted changes?
No. Cursor's deployment docs say deploy builds the pushed Git ref, so uncommitted changes don't ship. [3]
Can you self-host an Agent SDK project instead of using Cursor's managed hosting?
Yes. Cursor's deployment docs describe self-hosting with agent-sdk serve, and say to run the Agent SDK under Node 22.13 or newer, never Bun. [3]
Does the Agent SDK support Privacy Mode (Legacy)?
No. Cursor's deployment docs say deploy, a hosted session, and an eval run all refuse under Privacy Mode (Legacy); standard Privacy Mode is supported. [3]
Next step
Cursor's Agent SDK has no affiliate or paid partner program, so readers are sent to the official docs for setup details.
Read Cursor's Agent SDK documentation