AI generation and MCP

AI generation workflow

AI generation uses your server-side OpenRouter configuration. Keep the key out of browser bundles. Configure the model on the server and verify provider access before debugging the editor interface.

1. Save your current work.
2. Describe the page purpose, structure, content and responsive expectations.
3. Select the intended editing scope in the studio.
4. Review generated native layers and inspect the responsive result.
5. Save and preview; publish only after explicit review.

The client generate operation supports streamed events and cancellation. Treat a cancelled or failed run as incomplete and inspect the current document before retrying. Redis participates in AI rate limiting; provider limits and billing remain separate.

MCP discovery

Server configuration can register operator-owned MCP servers using stdio, HTTP or SSE transports. The backend exposes list and inspect operations for discovery. This integration is not unrestricted tool execution by a browser user.

Configure authorizeMcp to check the requesting user's authorization. The server must own command paths, arguments, environment variables and remote endpoints; do not accept arbitrary connection configurations from visitors.

Agent-driven editing

For a coding agent working from a terminal, use the CLI workflow. MCP discovery and CLI document editing are separate capabilities. Neither replaces host authentication or explicit approval to publish.