The MCP and CLI guide
Two ways for AI and developers to work on your website on Webmati.
Besides the admin, Webmati has two doors for AI and for people who write code:
- The MCP: a Model Context Protocol server at
https://mcp.webmati.com/mcp. Connect it to Claude, Cursor, Codex…, then chat: the AI uses the platform's tools to create pages, edit the theme, add products, build a backend, read your analytics. Nothing to install on your machine. - The CLI: the command-line tool
webmati. It copies a website onto your machine as a folder — theme, backend workflows, pages… — that you (or a coding agent) edit with your usual editor, check with the server's own rules, and push back to the drafts.
Which one?
| You want to… | Use |
|---|---|
| have an AI do one specific thing in a chat ("add an About page", "change the button colour") | MCP |
| work right inside the AI client you already use, nothing more to install | MCP |
| edit many theme files, write Go code for workflows, use git, review before you push | CLI |
| let Claude Code, Cursor or Codex work on your machine, spending fewer tokens | CLI |
| run a workflow without touching real data | CLI (webmati test) |
Both share the same data with the admin: a change made in one place shows in the
other two. Working over the MCP and want to move to your machine? Tell the AI "I
want to work on this website locally": it calls init_local_project and hands you
the commands to install the CLI, sign in and get the project — see
Work locally.
When the platform gets in the way
Something on Webmati refuses what it should accept, gives a wrong result, or
lacks what you need? Tell the platform team on Bugs & ideas, or
let your AI do it while it works: webmati issue new --attach-last from the CLI,
report_platform_issue over the MCP. Both look for the same report first so you can
+1 it instead of filing it twice, and you hear on your bell and by mail when it
is fixed or answered. The title and text are public to everyone signed in, so no
keys and no customer data; the logs and where it happened stay private to you
and the platform team.