How to add design review to Cursor
Add the TheDesignAgent MCP server to Cursor's mcp.json, locally or hosted, and Cursor's agent gets a design brief before it builds a screen and a scored UX and visual review after.
To add design review to Cursor, add the TheDesignAgent MCP server to Cursor's mcp.json. Cursor's agent then has a design agent in its loop: Discover gives it a brief before it builds a screen, and Ux and Visual score the result for job fit, UX and visual quality, using a project memory every agent on the repo shares.
Before you start
- A TheDesignAgent account. The hosted server signs you in through the browser; the local server needs an API key from your dashboard, which starts with
tda_. - For the local server, Node.js 20 or later.
Local or hosted?
- Local runs on your machine through
npx. All three tools work, andVisualcan screenshot pages onlocalhost. - Hosted needs no install.
Discover,UxandVisualall work, butVisualcan only reach public pages and preview deploys, notlocalhost.
Most people building locally want the local server.
Add the server
Open Cursor's MCP settings, or edit .cursor/mcp.json in your project (~/.cursor/mcp.json applies it to every project).
Local:
{
"mcpServers": {
"thedesignagent": {
"command": "npx",
"args": ["-y", "--package=@thedesignagent/mcp", "thedesignagent-mcp"],
"env": { "THEDESIGNAGENT_API_KEY": "tda_..." }
}
}
}
Hosted, with no key:
{
"mcpServers": {
"thedesignagent": {
"url": "https://www.thedesignagent.ai/mcp"
}
}
}
Cursor shows the server as needing a login. Click it and sign in to your TheDesignAgent account in the browser. To use a key instead, for example on a shared machine, add "headers": { "Authorization": "Bearer tda_..." }.
Keep the --package form for the local server: the package has two commands, so a bare npx @thedesignagent/mcp won't start.
Keep the key out of your repo. Run npx -y --package=@thedesignagent/mcp thedesignagent login once to save it to ~/.thedesignagent/credentials, then drop the env line from the local config. Don't commit a project mcp.json that contains a key.
Tell the agent when to use it
The tools tell Cursor's agent what they do, but a rule makes the loop reliable. Add a project rule along these lines:
For any new screen or significant UI change: call Discover with the task before building,
build from the brief, then call Ux and Visual (with render_url for a running page) and
fix critical and high findings before finishing.
Cursor can also install the plugin from its repository if your version supports Agent Plugins: https://github.com/CMBurnett/the-design-agent-plugin. That adds skills which teach the same loop.
Check it works
Ask Cursor to build a small screen. On a new project, the first Discover asks the agent to scan the repo and send a project model; after that, briefs come back immediately. Then watch for the Ux and Visual calls once the screen is built.
If something's wrong
If the tools don't appear, check the server shows as connected in Cursor's MCP settings and that the key is set. The support page covers the common error messages. Other clients: Claude Code and Grok Build and Codex.
Common questions
How do I add design review to Cursor?
Add the TheDesignAgent MCP server to Cursor's mcp.json, either the local server through npx or the hosted server at https://www.thedesignagent.ai/mcp (sign in through the browser, no key), then add a project rule telling the agent to call Discover before building and Ux and Visual after.
Should I use the local or hosted TheDesignAgent server in Cursor?
Use the local server if you build on localhost: all three tools work and Visual can screenshot local pages. The hosted server needs no install but Visual can only reach public pages and preview deploys.