Which tool built the site
The app notices which tool built a project from the files tools leave behind, and the project's "Your AI" card says when Claude Code is working in the folder right now, with a live read-only view of the session.
Starting your AI in the right folder
Start Claude Code here (or Codex, whichever you chose) opens a terminal in the project folder with the tool already running. The AI page shows which tools are on this computer, which build, whether each is signed in, and the one command that fixes whichever of those is wrong — and the app runs that command for you.
The /deploy command
For Claude Code the app writes a /deploy slash command into the project. Type it in a Claude Code session and the app publishes the project, through the app, with the hosting credential still in the app: the AI never sees it, and it can never skip the key check. The list of projects that have the command, and the ones that still need it, is on the AI page.
The AI built in
The app has an AI of its own, V&D.AI, with nothing to install and nothing to sign into but your account. Ask for a change in your own words — "make the heading bigger and the button green" — and it reads the files in that folder and edits them, in a window you can close and come back to. A project holds several conversations. A line starting with / is for the app rather than the model: /deploy publishes without leaving the chat, /model says what it runs on.
It is billed to a balance in dollars, because that is what it costs us; a new account starts with a dollar from us. More about it →
Standing instructions
The AI page has a box for the instructions every new site should carry — colours, tone, language, contact details, anything you would otherwise repeat. They travel with a project the app asks an AI to build; a chat started elsewhere does not inherit them.
Connecting a tool over MCP
The app has an MCP server, which is how an AI tool gets tools of its own. On the AI page, under Inside your AI, each tool on this computer has one press:
- Claude Code and Codex are registered through their own commands (
claude mcp add,codex mcp add), which the app runs for you, in the user scope so the server follows you into every folder. - Cursor gets one entry in its
~/.cursor/mcp.json, with everything else in the file untouched.
Disconnect undoes the same. Whether a tool is connected is asked of the tool each time the page opens, not remembered. The server runs under your own Node.js, so the row says so if there is none.
Once connected, the tool can publish, and — for a local site — see what it is, start and stop it, run WP-CLI, read its logs, set its config, query its database and sign in to its dashboard. Nothing it can do deletes a project, a database or an image; those stay in the app. Every tool, with what it takes and answers →
For any other MCP client, the line under the rows carries the server's path; point the client at node <that path>.