Back to Blog
Cursor Live Apps: What It Deploys and What You Still Own
Cursor's November 5 Live Apps release can scaffold, host, and share a web app from a cloud agent. The official docs are clearer than the changelog about what the publish checks cover, and what they leave to you.
Back
Engineering
Table of contents
On November 5, 2026, Cursor shipped Live Apps in its changelog, called Apps in the docs. You describe an internal tool, and a cloud agent builds it, stands up a database if needed, puts sign-in in front of it, and gives you a link. I already run sites on a VPS, so I wanted to know what that actually covers.
This is a reading of the official changelog and the current Apps docs, not a hands-on publish of my own. Where they disagree, I follow the docs. Some important details are not written down yet.
What you actually get
You have to use a cloud agent. The docs say this is not supported from a local agent, and you have to say plainly that you want Cursor to build or deploy an app.
Start from scratch, and Cursor creates a new repository on Cursor Origin. Or ask the agent to deploy an app from an existing GitHub repository. The November 5 changelog said GitLab support was coming soon. The current Apps docs still only mention Origin and GitHub, so I would not assume GitLab works yet.
When the agent is ready, you get a tray to preview the app and publish it. Publishing is a separate, explicit step. You pick the app's URL, which you can change later, and Cursor runs security checks. If they pass, you get a live link. If they fail, it lists what needs fixing, you tell the agent to fix it, and you publish again. Nothing goes live until you choose to publish.
Cursor then deploys the app in a dedicated hosting account it creates for your team, or for you on an individual plan, isolated from other teams and users. If the app needs a database, it provisions Postgres automatically. Private and team apps sit behind Cursor sign-in. If the app needs to reach your own systems, such as a data warehouse, Cursor prompts you for the secrets, and the docs say the agent never sees those values. Team admins usually create team secrets in the Apps dashboard, and you can add personal secrets if you have access.
The changelog calls that sign-in Cursor SSO. The Apps docs call it Cursor sign-in (Cursor's own login). Only people you share with, who are signed into Cursor, can open a private or team app. The Apps docs do not describe a separate SAML setup just for the published app.
Visibility is one of three options: people you invite, the whole Cursor team, or public (anyone with the link). Public sharing is available on individual plans. On Teams and Enterprise it is off by default. If an admin later disallows public apps, existing public apps drop back to team visibility.
Scripts cannot pass that sign-in on their own. Cursor documents a workaround: send a Cursor API key in the x-cursor-app-token header on every request. Admins can allow or disallow this.
Cloud agent setup also has a separate Publish flow that connects a Vercel account and deploys the default branch. That path needs a Vercel account and is not the Live Apps publish button. Live Apps, on the Apps page, is Cursor provisioning its own isolated hosting account, plus the checks and sharing model above.
What the guardrails cover
Four checks run before every publish. The docs say they are always on, even if the AI security review is turned off:
- Block secrets written in app files (API keys or access tokens).
- Block credential files (private keys, service-account keys, or files with auth tokens).
- Never publish
.envfiles. They are left out of what ships. - Never publish source maps, so visitors cannot read the original source.
An AI security review then reads the app's files and, in the docs' words, blocks serious issues before the app goes live. Team admins can add instructions to that review prompt. Those instructions can add rules. They cannot remove Cursor's built-in checks.
Team admins can see every app and every secret on the team, unpublish any app, and turn Apps on or off for the team. They can approve a publish that the review blocked, except when the block is for malicious code, hardcoded secrets, or exposed credentials. Changes to these settings go into the team's audit log.
That is a real gate, but not a full review of the running app. The always-on checks are about what gets shipped in the files. The AI review is described only as reading those files and blocking serious issues. The docs do not list those issues, and they do not claim the scanner will catch every secret or every vulnerability.
What still sits with you
The official docs leave several gaps.
You still decide which systems the app may call, who may create secrets, and how those secrets are rotated.
If the app needs a database, Cursor provisions Postgres. The docs do not say who operates that database, which region it lives in, how large it can get, whether you get a direct connection, how backups work, or how you export the data if you leave.
You pick a unique domain, and you can change it later. The legal text on the Apps page talks about a subdomain. I could not find any official mention of pointing your own custom domain at a Live App.
Cloud agents bill against your plan's usage at the selected model's rate, and you set a spend limit when you first start using them. They require a plan that includes cloud agents: Premium and higher, Teams, Enterprise, or a legacy paid plan. Free and Plus do not. Apps is available on those same paid plans. I could not find published figures for hosting compute, bandwidth, storage, request limits, or a separate price for the Postgres instance. The Apps page says hosting is provided as is and as available, and that a subdomain, account, or published app may be revoked or suspended. There is no documented uptime target.
Starting from scratch puts the code in a new Cursor Origin repository, which you can browse and clone. Starting from GitHub keeps the code where it already is. The hosted app and the provisioned database sit in an account Cursor creates for you. The docs do not describe a one-click export onto your own VPS. Moving off later means taking the repo, standing up your own database, and re-pointing clients.
Who it fits
This looks useful for small internal tools I would otherwise leave on localhost: a dashboard, a signup form, a status board, a portal that only people already on your Cursor team should see. It is a poor fit if you need a custom domain, a database you can back up and inspect yourself, a public site on a Teams plan whose admin has not allowed public apps, or a production service you intend to keep on your own VPS.
If you try it, treat the publish button as the real decision, not the scaffold. Read the failed checks, keep secrets out of the repo even though the scanner is there, and do not put data in the provisioned Postgres until you are comfortable with how little is documented about it.
Sources
Comments
No comments yet
Loading comments...