On this page

Cursor has shipped the Agent SDK as generally available. The pitch is agent-as-code: put the bot in a project folder, test it with evals, then deploy it to Cursor or run it on self-hosted infrastructure.

What shipped

Per the Agent SDK changelog, the product “is available today” and “generally available today.” An agent is a project. instructions.md is 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. agent-sdk validate checks the project; agent-sdk info prints what was discovered.

Deploy with agent-sdk deploy from the project directory. Cursor says the command builds the agent, gives it a URL, and starts it on Cursor’s infrastructure; self-hosted infrastructure is also a target. Secrets and integrations are managed from the dashboard, with the same logs, evals, and playground available locally.

Install—and the package-name mismatch

Docs and npm list a global install of @grok/agent-sdk, then scaffold and a local playground:

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

As of Beijing time October 27, 2026, npm lists @grok/agent-sdk at 0.3.0. The same version ships under the former name @cursor/bdk. The changelog’s starter line uses npx @cursor/agent-sdk init ./my-bot—that package name returns 404 on npm. Install commands use the docs／npm name @grok/agent-sdk.

Pass --template on init to start from a security reviewer, an owners-policy approver, a PR autofixer, or a Linear or Jira triager.

Channels, evals, and replay

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. Cursor says GitHub events pull through a Cursor account, so no public webhook URL is required for that path. Each conversation keeps its own session; follow-ups on a pull request or Slack thread resume the same session.

agent-sdk eval sends a fixed message and asserts on which tools ran, with what input, and the shape of the reply—locally or in CI. agent-sdk github replay turns a real pull request into signed webhook payloads so GitHub-driven bots can be tested without waiting for a push; --dry-run --fixtures-dir can save those payloads as eval fixtures.

Deploy, eval, and replay have not been run end-to-end on this site—for reference only. Cursor’s docs also compare Agent SDK with Cursor Automations and Grok Bot in a product table (versioned Markdown／TypeScript project vs dashboard prompt／trigger vs app-configured agent); that framing is Cursor’s, not an independent ranking.

Related reading: How Cursor Usage Limits Work on The X Subscription and Legacy Plans (October 2026), How Claude Code and Cursor Compare on Price and Limits (October 2026), and Best Coding Agents Right Now, Ranked by Artificial Analysis (October 2026).