Agents handle everything from building to publishing
Cursor, the AI coding tool, on Nov. 5 (Nov. 6 Japan time) added a feature that lets AI agents build web apps that can be published and shared by link. The company said on X that if users describe the tool they want, Cursor will build it, set up a database, run 2 types of security checks, one based on fixed rules and one by an agent, and carry it through to deployment (publishing).
You can now build live apps in Cursor and share them with a link. Agents have made it easier to write code, but most apps never make it off localhost. Describe the tool you want - Cursor will build it, set up a database, run deterministic and agentic security checks, and deploy.
In its changelog, the company explained the aim of the feature by saying that many apps and tools built by users never leave their own computers. Examples of uses it cited include operations tools, customer-facing explainers, onboarding for new hires, sales pipeline dashboards and service catalogs for engineering departments.
Databases and sign-in set up automatically
Users start an agent running in the cloud and ask it to build an app. The agent creates a scaffold, provisions a Postgres database if needed, adds sign-in with Cursor accounts and secures a place to host it. When connecting to internal data infrastructure or other systems, it asks the user for secrets such as the keys needed to connect. According to the documentation, the agent does not see those values.
New apps are stored in "Cursor Origin," Cursor's code repository, and existing apps can be published from GitHub. Support for GitLab and others is coming "soon," the company said. Apps run on dedicated hosting separated by team or individual.
Checks before publishing, and administrators can block it at companies
Checks run at publishing time, and if problems are found, the agent points out what to fix. Nothing is published until the user chooses to publish, and the user sets the app's URL when publishing.
According to the documentation, 4 checks are always run. They block API keys and access tokens written into the app's files, block credential files such as private keys, and do not publish ".env" configuration files or source maps that would expose the original source code. In addition, an AI security review reads the files and stops publishing if serious problems are found.
Team administrators can decide whether to enable the feature and whether to allow publishing externally or use via API keys. They can also approve publishing that was stopped by the review, but not when it was stopped for malicious code, embedded secrets or leaked credentials. Administrators can see all of the team's apps and secrets and can stop any app from being published. Changes to settings are recorded in audit logs.
The company does not, however, say that apps passing the checks are free of vulnerabilities. The documentation says Cursor only provides technology and tools, offers no guarantee and provides hosting "as is." No external evaluation measuring how many problems the checks catch has been found.
Sharing within teams by default
As a rule, apps sit behind Cursor sign-in and can be shared with invited people or groups or the whole team. On individual plans, apps can be published so that anyone with the link can see them. On Teams and Enterprise plans, external publishing is disabled by default and can be enabled by administrators.
By issuing a Cursor API key, users can access apps from programs or from "Grok Bot," an AI agent whose development the company takes part in.
Paid plans only; pricing and limits not disclosed
The feature is offered on all paid plans with access to cloud agents. According to the cloud agent documentation, these are Premium and above, Teams, Enterprise and legacy paid plans, and do not include Free and Plus. Agent work is deducted from the plan's usage allowance at the unit price of the selected model, and users set a spending limit the first time they use it.
Neither the changelog nor the documentation states hosting fees for published apps, database capacity, usage limits or data retention periods. AlphaSignal noted that apps requiring a different network configuration or authentication system still need their own publishing process, as before.
An extension of a series of feature additions
Over the past 2 weeks, Cursor has added in quick succession "Computer Use," in which agents operate desktop apps to test features they built themselves; general availability of "Agent SDK," a development kit for building one's own agents; and "Annotate," which lets users mark up a running screen to have it fixed. With the new feature, users can leave it to agents to carry what they build through to the stage where others use it. Replit and Sweden's Lovable offer development services that handle everything from database setup to publishing.
At the end of October, Cursor disclosed a problem in which tampered extensions were distributed through its extension store. For a feature that publishes apps built by employees, how far the pre-publishing checks and administrator permissions work will be the focus for corporate adoption.
