Cursor can put your app online. Who gets through the door?
The November 5 release combines building, hosting and sign-in. Its sharing defaults and publication controls matter as much as the generated code.
Editorial visualizationA small internal tool no longer has to end as code on a laptop. In its November 5 release, Cursor added a route from a request to a hosted web app: its cloud agent can build the application, provision a Postgres database when needed, configure authentication and provide a shareable address. The release is available on paid plans with cloud agents enabled.
That makes the audience decision immediate. A working link may be intended for named colleagues, a whole team or anyone who receives it. Choosing that boundary is part of finishing the app.
To start, explicitly ask a cloud agent to build an app. Cursor says it can deploy from new or existing repositories on Cursor Origin and GitHub; GitLab deployment support is still forthcoming in the launch announcement. The workflow prepares the app, then requires an explicit publication action. It is not a promise that every generated project is ready for customers.
A public link is a separate choice
The current Apps documentation distinguishes these audiences. Public sharing is disabled by default for Teams and Enterprise, but an administrator can enable it. Individual plans also support public sharing. These are access settings, not an assessment of the information an app exposes.
| Visibility | Who can open the app |
|---|---|
| Invited people | The owner or selected team members and groups, signed into Cursor. |
| Team | Members of the Cursor team, signed into Cursor. |
| Public | Anyone with the link; Teams and Enterprise require admin enablement. |
Editorial visualizationFor a first trial, a useful question is whether the app still works with deliberately uninteresting sample data. A prototype that displays invented signups can demonstrate a filter without exposing a customer list. Before widening access, try the app as an intended viewer rather than only as its creator.
What happens before publication
Cursor's publication controls block embedded secrets and credential files, exclude .env files and source maps, and add an AI security review. Administrators can supply extra review instructions. They can override some blocked publishes, but the documented exceptions exclude malicious code, hardcoded secrets and exposed credentials. Admins can also unpublish apps.
Those checks address recognizable publication hazards. They do not establish that a calculation is correct, that a form handles every failure, or that the app meets a team's operational requirements. A passing review and a successful business task are different tests.
Editorial visualizationThe underlying cloud workflow also deserves attention. Cursor's security overview says each agent runs in an isolated virtual machine and inherits the initiating user's repository access. It also warns that agents automatically execute terminal commands: malicious instructions in material they read could attempt to send code elsewhere. Network restrictions and carefully scoped access remain relevant while building, before a publication scan ever runs.
For teams assessing data handling, that overview distinguishes temporary runtime workspaces from saved conversations. It describes VM snapshots expiring after 30 days of inactivity, while conversation records persist by default until deleted or limited through available retention controls. Building in the cloud therefore involves more than deciding who sees the finished page.
This account documents Cursor's stated behavior; UpThat has not independently tested deployment reliability or the security review's detection rate. A sensible first release is a bounded tool with a named owner, a clear audience and an observable task to test. If a colleague can open it, complete that task and obtain the right result, there is something concrete to evaluate beyond the live link.
Publication history
— First published.
No published corrections or earlier versions.
How this was made
UpThat uses AI-assisted research and writing. Source checking and publication follow our editorial review workflow. See the original documents above to assess the reporting.
How we use AI ↗

