Skip to main content
Ask your agent to bring a screen or component from your code onto the canvas. It reads your code and rebuilds what it renders as editable layers, with your own colors, fonts and icons, so you design from what your app really looks like. Before you start · Desktop app or any MCP client · Free plan In the desktop app, you need a repository and a chat with Claude Code or Codex. See Add a repository. With another coding agent, such as Cursor, you need the Wonder MCP installed in it and your repository on your computer.

Ask for a screen or component

  1. Open a chat in a worktree of your repository. With another coding agent, open your repository in it.
  2. Say what to bring, using the names your code uses, for example:
    or
  3. Send it.
If the worktree has no Wonder file yet, your agent asks Where should I work on the canvas? and offers to create a file for the worktree. Choose that, or add a file yourself: see Give a worktree its design files. Your agent works from your source code, not from your running app:
  • It finds the parts of the screen in your code, and keeps every control, row, text, icon and layout of the state you asked for. It doesn’t add UI your code doesn’t have.
  • It builds the screen section by section, checks each against your code with screenshots and fixes what differs.
  • When it’s done, it says what it reproduced, which repository and branch it read, and anything it couldn’t match, such as a value it couldn’t read or an image it couldn’t bring in.
To compare with your running app, ask your agent to check the result against the Preview.
A worktree chat where the agent reports the home page it brought into the Gadwall file as the Home Desktop Rose artboard, with its finished todo list, next to the canvas showing that artboard selected and its layers in the Layers panelA worktree chat where the agent reports the home page it brought into the Gadwall file as the Home Desktop Rose artboard, with its finished todo list, next to the canvas showing that artboard selected and its layers in the Layers panel

Use your variables and icons

You don’t need to ask for these separately: a screen built from your code brings in what it uses.
  • Colors, sizes and fonts use your file’s variables. A value your file doesn’t have yet becomes a new variable with your code’s name and exact value. If a variable has the same name as one in your code but a different value, your agent keeps yours and tells you. See Bring in your code’s variables.
  • Icons come from your repository’s own icon files and join the file’s icon library. Your agent adds only the icons the screen uses, not your whole set. See Bring in your code’s icons.
  • Fonts are the ones your app uses.
  • Images come from your repository. Your agent doesn’t swap in look-alikes: an image it can’t bring in shows your code’s own fallback, such as initials or a placeholder, or stays empty when there’s none, and it tells you.

Refresh a design after the code changes

When your code changes, ask your agent to bring the screen in again, for example:
Your agent reads the current code, updates the design and its variables to match, and says what changed. To keep the old version, ask it to work on a copy, for example “refresh it as a copy next to the old one”. To make either choice your default, save it as a design preference.

Import your component library as a Design Guide

To bring in all your components at once, ask for the whole library:
This takes 20–30 minutes or more for a typical codebase, so give it a scope when you can. Your agent:
  1. Lists your components from your Storybook stories and your component folders.
  2. Brings in your variables and icons first, then the basic components, then the components built from them.
  3. Puts them in one frame named … Design Guide, with a Guide Section per component. Each state your code defines, such as hover or disabled, gets its own layer, and a repository with a dark theme gets a light and a dark version of each.
To keep it shorter, name what to bring, such as “our buttons and forms” or packages/ui. On a large repository with no scope, your agent starts with your variables, icons and basic components, says what it left for later and offers to continue. It builds full screens from the library only when you ask. On the Free plan, a large import can use a good part of your weekly AI-generated layers.

Other ways onto the canvas

  • Capture your running app: turn what the Preview shows into editable layers, when you want the screen as it runs, with real data.
  • Wonder Capture: capture any webpage, or part of one, from your browser.
  • Wonder MCP: use Wonder from Cursor, Claude Code or Codex outside the desktop app.

If something goes wrong

  • Your agent asks “Where should I work on the canvas?”. The worktree has no Wonder file. Let it create one, or add one.
  • Your agent asks where your code is. Your coding agent can’t see a copy of your repository. Open the repository’s folder in it, or work in a chat in the desktop app.
  • Some values are listed as unverified. Your agent couldn’t read them, often because a package isn’t installed. It used the closest value it could find. Install your dependencies and ask it to check again.
  • An image or logo is empty. Your agent couldn’t find it in your repository. Add it to the design yourself, or tell your agent where it is.
  • The design doesn’t match your app. Tell your agent what’s different, or ask it to compare with the Preview.
  • “Your agents have made 3,000 layers this week on the Free plan.” See Plans and limits.