Get started
View as markdown.mdMCP server
Let your AI assistant manage your Zeitlos project — tables, fields and settings — while it writes your code.
Zeitlos exposes an MCP server, so an AI assistant can manage your project directly: create tables, add fields, define profile fields and read your project's configuration.
Connect
The server is at:
https://mcp.zeitlos.app/mcpAdd it to your assistant. For Claude Code:
claude mcp add --transport http zeitlos https://mcp.zeitlos.app/mcpOr in an .mcp.json file:
{
"mcpServers": {
"zeitlos": {
"type": "http",
"url": "https://mcp.zeitlos.app/mcp"
}
}
}You'll need to authenticate the MCP before using it, e.g. running /mcp in Claude.
What it can do
| Area | Examples |
|---|---|
| Database | List tables, describe a table, create tables, add fields, query and create records |
| Authentication | List and add profile fields, read auth settings |
| Project | Read project details and configuration |
Tools are gated on what the project actually has — database tools appear once the database is enabled, auth tools once authentication is.
A typical flow
Ask for the schema you need in plain language:
Prompt example
Create a posts table with a title text field, a published boolean and a author
relation to the users table.
Then write the code against it. Because the assistant created the schema, the code it writes and the database it writes to stay in sync.
Tip
The @zeitlosapp/sdk package ships an AGENTS.md alongside it. If your assistant reads
the package, it picks up the SDK conventions without you explaining them.
Local development
npx @zeitlosapp/cli dev runs a local Zeitlos emulator — database, auth and its own MCP server — on
your machine. It prints the local MCP port on startup:
claude mcp add --transport http zeitlos-local http://localhost:7312/mcpThe local server announces itself as the local project, so an assistant connected to both this and the production server won't mix them up.
Because schema lives in the project and not in your repository, a local project starts empty. Either ask your assistant to create the tables your code needs, or have it mirror an existing cloud project:
Prompt example
Using the production Zeitlos MCP, list the tables in project my-app and describe each
one. Then recreate that exact schema in the local database via the zeitlos-local MCP.
That copies schema, not data — ask for specific rows too if you want sample data.
See Local development for the rest of the local database workflow, including generating types.