You need a Wonder account with at least one file. The MCP server is free on every plan, including the free tier.
What is Wonder MCP?
Wonder’s MCP server plugs into coding agents (Claude, Cursor, and others) so that designs created in Wonder can be read, edited, and updated by your agent, and vice versa.How it works
Wonder connects to your agent through the Wonder MCP server athttps://mcp.wonder.so/mcp. On first use you’ll sign in through a standard OAuth flow. There are no API keys to manage, and tokens refresh automatically.
For Cursor, install Wonder from Cursor’s plugin marketplace. Claude Code and Codex support the Wonder plugin directly.
Install
Pick your agent below. We currently support Cursor, Claude Code, Claude Desktop, Claude custom connectors, Codex, Antigravity, Lovable, and custom MCP clients.
Cursor
Open Customize
- Click Browse Marketplace.
- Search for Wonder and open the result.
- Click Add.
- Click Manage. Wonder should now appear under MCP.
- Click Authenticate and complete the browser sign-in.

Claude Code
Plugin (recommended)
Open your terminal and run claude to start Claude Code. Add the marketplace:

Claude Code showing the Wonder plugin available to install
/plugin and confirm that Wonder appears under installed.

Claude Code showing the Wonder plugin under Installed
Manual setup
- Open the terminal on your computer.
- Run the command below (not inside Claude) to register the Wonder MCP server:

Run the command in your terminal
- Run
claude mcp listto confirm the Wonder server appears.

Confirm the Wonder server appears
- Open a Claude Code session and run
/mcp. You should see Wonder in the list of MCPs. If you don’t, restart your Claude session. Wonder will prompt you to sign in to your account in the browser. Once done, your agent is connected. Tokens refresh automatically, so you should only need to do this once.

Wonder appears in the /mcp list

Claude Desktop
Plugin (recommended)
- Open your Claude Desktop app and navigate to the Code tab in the top-left corner.
- Click on Customize.
- Click the plus icon under Personal Plugins > Create Plugin > Add Marketplace.
- Enter the Wonder plugin URL in the field:
aquila-lab/wonder-plugins. - Click Sync and add the Wonder plugin that appears to Claude.

Add Marketplace in Claude Desktop

Wonder plugin installed in Claude Desktop
claude session in your terminal and run /mcp.
Manual setup Claude Desktop requires Node.js to run MCP servers. Verify by running
node --version in your terminal.
Open Claude Desktop Settings > Developer > Edit Config. This opens your claude_desktop_config.json file.
Your file probably looks something like this:
} of mcpServers. Fully quit and restart Claude Desktop. A browser window will open asking “MCP CLI Proxy would like access to your account.” Click Allow access. Once the browser confirms “Authorization successful!”, close it and return to Claude. Check the chat input area for the MCP indicator to verify Wonder is connected.

Claude (custom connector)
If you use the Claude desktop app or claude.ai in the browser and don’t want to touch the terminal or config files, add Wonder as a custom connector. This is the simplest way to connect Claude to your canvas.
There are two parts. An admin adds Wonder as a connector once, then each person connects their own Wonder account. On a personal Pro or Max plan you’re the admin, so do both.
Add Wonder (admins)
- Open Claude (the desktop app or claude.ai).
- Go to Settings > Connectors.
- Click Add, choose Custom, then Web. In some versions this reads Add custom connector.

Add a custom connector from the Connectors settings
- Enter the following details:
- Name:
Wonder - Remote MCP server URL:
https://mcp.wonder.so/mcp
- Name:
- Claude auto-detects Wonder’s authentication settings. Keep the detected defaults:
- Authentication: Always required
- OAuth client: Use Anthropic’s hosted client metadata (Recommended)

Keep the detected authentication defaults
- Scroll down and click Add to make Wonder available.
On Team and Enterprise, only a Claude organization Owner or Primary Owner can add a custom connector. If you don’t see the option, you likely aren’t an Owner or your organization has restricted custom connectors: send your admin this guide.
Connect your account (everyone)
- In Claude, open Customize > Connectors.
- Find Wonder in the list (it may have a Custom label) and click Connect.
- Sign in to Wonder and approve the connection.
- Open the Wonder connector settings. Under Tool permissions, set both groups to Always allow:
- Read-only tools: Always allow
- Write/delete tools: Always allow

Set both tool groups to Always allow
This lets Claude work without asking before every Wonder tool call. Claude can still only reach files your Wonder account permits.
- Open the Wonder web or desktop app and keep the canvas file you want to use open.
- In a Claude conversation, open + > Connectors and enable Wonder if it isn’t already.

Codex
Plugin (recommended)
- With the Codex CLI installed, run these commands in your terminal:
- Restart Codex and start a new chat. Type
@wonderand select the installed Wonder plugin.

Select the installed Wonder plugin
The Wonder · Computer use entry starts desktop automation. Select the installed Wonder plugin to use its MCP integration.
- Open a canvas in Wonder, then send this read-only connection check in Codex:
- Complete Wonder sign-in when Codex prompts you, then retry the connection check. Confirm Codex calls a Wonder MCP tool and returns your canvas’s artboards before asking it to edit a design.
Manual setup Run the command below in your terminal to register the Wonder MCP server:
config.toml. Run codex mcp list to confirm the Wonder server appears.
Restart Codex and start a new chat, then run the read-only connection check above and complete Wonder sign-in if prompted. Manual setup connects the MCP server without the plugin’s bundled skill, so ask explicitly to use Wonder MCP.

Antigravity
In Antigravity, click the ⋯ menu in the Agent panel > MCP Servers, then select Manage MCP Servers > View raw config. This opens your mcp_config.json file.
The file is located at:
- macOS / Linux:
~/.gemini/antigravity/mcp_config.json - Windows:
C:\Users\<USERNAME>\.gemini\antigravity\mcp_config.json
serverUrl instead of url for remote MCP servers.

Lovable
Custom MCP servers require a paid Lovable plan.
- Open Lovable.
- Navigate to Settings → Connectors → Personal connectors via the sidebar. Scroll down to New MCP server and click it. You can also click + in the prompt box and go to Connectors for quick access.

Add Wonder to Lovable with a custom MCP server
- Enter
Wonderas the MCP name and paste this URL:
- Select No authentication and click Add server.
Other MCP clients
If your client isn’t listed above, you can set up Wonder manually. The server URL is:mcp-remote bridge (requires Node.js):
Try it out
Once Wonder is connected, open a chat in your agent of choice and try one of these prompts to confirm everything is working. If you are in an existing codebase:Troubleshooting & FAQ
What does the Wonder MCP server do?
What does the Wonder MCP server do?
Do I need a Wonder account?
Do I need a Wonder account?
Is the MCP server free?
Is the MCP server free?
Does it work with Windsurf, VS Code Copilot, or Cline?
Does it work with Windsurf, VS Code Copilot, or Cline?
"url": "https://mcp.wonder.so/mcp" works with most clients. For clients that don’t support remote URLs directly, use the mcp-remote bridge method shown in the Other MCP clients section.The Wonder server isn't showing up, or tools aren't working
The Wonder server isn't showing up, or tools aren't working
The MCP server is connected but the agent isn't using it
The MCP server is connected but the agent isn't using it
Why does Codex still use Computer Use?
Why does Codex still use Computer Use?
Can the agent modify my designs?
Can the agent modify my designs?
I already added other MCP servers. Will Wonder conflict?
I already added other MCP servers. Will Wonder conflict?
mcpServers entries in the same JSON object:Where can I report issues or request features?
Where can I report issues or request features?
Related
- Bring your code onto the canvas: connect a GitHub repository and design with your real components.