Skip to main content
Your agent runs on a cloud sandbox with your workspace’s code, writes the app, and publishes it. No local setup on your end.
A workspace is a Zite database plus the apps built on it. See Workspaces.

​
1. Connect the Zite MCP

Add the Zite MCP server to your AI tool and authorize it. Claude and ChatGPT have native Zite integrations in their app directories. Everywhere else, use the MCP URL:
On first connect you sign in and authorize access. The agent then acts as you, so it can only see and change workspaces you have access to. Using VS Code, Codex, Windsurf, Zed, or another client? See the per-tool steps in Connect.
Building over MCP is free for 14 days, starting the first time your agent boots a sandbox. After that the build tools are on the Business plan and above.

​
2. Describe what you want to build

Tell the agent the system you want. Be specific about the data and the screens:
The agent spins up the workspace (a fresh database) and a cloud sandbox, then designs the tables, scaffolds the app, and writes the React UI and backend workflows. It iterates until the app type-checks and runs. That’s the build loop, which you can watch live in the editor. Already have a workspace? Name it in your prompt and the agent builds into it instead of creating one.

​
3. Choose who can access it

New apps are internal by default, so only your Zite organization can open them. To make an app public, ask the agent to set it external, which adds a sign-in screen (email, Google, or SSO). See Authentication.

​
4. Publish

External apps go live at https://<subdomain>.zite.so; internal apps become available to your team. Add a custom domain or roll back any time. See Publishing. Next: how building works and the database client.