The shortest path through 1DesignTool is one brief and one steer — about five minutes. This page walks that loop: a new project, a first run, and the first fix, so the rest of the docs have a shape to hang on.
Create the project
From the home screen, hit New project — or describe the design right in the home composer and the app creates the project for you. Picking a template tells the agent the output format (a slide deck, a mobile screen set, a UI mockup) and loads the right viewer and checks; Blank leaves the format to the brief.
Write the first brief
A good first brief carries the idea, the audience and the feel — one or two sentences, not a spec. "A landing page for a coffee subscription — dark, editorial, confident type" is enough. Send it with Send and the run starts: the composer shows the agent working, and the preview panel fills in as real files land in the project folder.
When the turn finishes, the gate reports — Design checks green means the result passes the kit's rules (contrast, type sizes, hit targets, working nav at every width). The result stays live in the right panel: hover it, click through it, drag the width.
Steer it once
Don't like a headline? ⌥-click it in the preview, write one line — "tighten this, two lines max" — and send. The comment lands as a pin the agent reads, and the next turn fixes exactly that spot. For bigger direction changes, write a follow-up in the composer instead.
Tip: Steer the what, not the how. "More space above the fold" beats "add
margin-top: 4rem" — the design kit translates your direction into tokens the whole project keeps consistent.
Take it further: a mobile app, re-designed over MCP
The same loop scales to bigger artifacts. Here a Mobile app project builds a habit-tracking app — Today rings, streaks, a stats screen — in native SwiftUI styling, then an MCP client (Cursor) drives the next turn from outside the app: a re-design that shifts the accent, bolds the rings and adds a weekly chart.
The external turn lands like any other — tagged via MCP · Cursor on the bubble, the run live in the transcript, and the gate reporting when it finishes. The studio never pauses; the client just uses its real API.
MCP — drive the studio from any agent covers the server, the access panel and what a client can call.