Working with AI agents (MCP)

MCP is a common language between AI apps (Claude Desktop, Claude Code, Cursor, and others) and services like Weblooks. Instead of copying and pasting between a chat window and your project, the assistant gets a set of Weblooks tools it can call itself, with your permission and inside your account's access rules.

Once connected, an agent can:

• list and open your projects
• create and edit pages and components
• read and write CMS collections and entries
• run scans and read the results, including alt text and metadata fixes
• deploy to staging and check deploy readiness

Everything runs under your account, so an agent can never do more than the signed-in person could.

In the desktop app, go to Settings → AI assistant connection and click Connect & verify. Weblooks detects the AI apps installed on your machine, writes the connection into each one, and verifies it end to end. One connection covers both your local folders and your sync projects. Restart the AI app and the tools appear.

Using Claude on the web, or a client on another machine? Add the cloud server by URL instead: https://app.weblooks.app/api/mcp, then approve access in your browser. The cloud connection reaches sync projects only, since it has no access to files on your disk.

For automation that signs in with a key instead of a person (a CI job or a GitHub Action, for example), create an API token under Settings → API tokens. Give it a name, choose read, write, or admin scope, and set an expiry. The token is shown once; treat it like a password, and revoke it from the same list if it leaks. For interactive AI apps, prefer the assistant connection above.