MCP Server

The Cortexcn MCP server lets an AI assistant search the registry, read an item's docs and source, and get the exact command to install it into your project.

Quick Start

1. Connect the server. In Claude Code, run:

   claude mcp add --transport http cortexcn https://ui.cortexcn.dev/mcp

   For other assistants, see Connect your assistant.

2. Make sure your project has shadcn set up. If there is no components.json in the project root, run:

   npx shadcn@latest init

3. Ask your assistant for what you need, for example: "Add the Lanyard 3D component to the about page."

About the server

The server is hosted, so most clients connect to the URL directly and there is nothing to install or run.

How installs work

The server never runs commands on your machine. When you ask for an item, it returns the install command, and your assistant (Claude Code, Cursor, or another agent with terminal access) runs it in your project root:

npx shadcn@latest add https://ui.cortexcn.dev/r/lanyard-3d.json

The command uses the item's URL, so you do not need to add the @cortexcn registry to components.json. The project still needs shadcn initialized. The CLI installs the item's npm dependencies and any other registry items it needs.

Chat apps such as claude.ai and ChatGPT cannot reach your terminal, so there the assistant gives you the command to run yourself.

Tools

Argument values:

• kind is one of component, block, chart, hook, or lib.
• limit defaults to 8 results, up to 25.
• include_docs defaults to true.
• package_manager is one of pnpm, npm, yarn, or bun, and defaults to pnpm.

Connect your assistant

Each client below uses the same URL. If a config file already lists other servers, add the cortexcn entry next to them.

Claude Code

claude mcp add --transport http cortexcn https://ui.cortexcn.dev/mcp

Add --scope user to use it in every project:

claude mcp add --transport http --scope user cortexcn https://ui.cortexcn.dev/mcp

Or add --scope project to write it to .mcp.json in the project root, so everyone on your team gets it. You can also write that file by hand:

{
  "mcpServers": {
    "cortexcn": {
      "type": "http",
      "url": "https://ui.cortexcn.dev/mcp"
    }
  }
}

Run /mcp inside Claude Code to check that the server is connected.

Claude Desktop and claude.ai

1. Open Customize > Connectors.
2. Choose Add custom connector.
3. Paste https://ui.cortexcn.dev/mcp as the URL and choose No sign-in.

In Claude Desktop you can use the config file instead. This runs mcp-remote through npx as a local bridge to the server, so it needs Node.js:

{
  "mcpServers": {
    "cortexcn": {
      "command": "npx",
      "args": ["-y", "mcp-remote", "https://ui.cortexcn.dev/mcp"]
    }
  }
}

The file is at:

• macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
• Windows: %APPDATA%\Claude\claude_desktop_config.json

Restart Claude Desktop after you save it.

Cursor

Add to Cursor

Or add it to .cursor/mcp.json in your project, or to ~/.cursor/mcp.json to use it in every project:

{
  "mcpServers": {
    "cortexcn": {
      "url": "https://ui.cortexcn.dev/mcp"
    }
  }
}

VS Code

The server works with GitHub Copilot in agent mode. Add it to .vscode/mcp.json in your project:

{
  "servers": {
    "cortexcn": {
      "type": "http",
      "url": "https://ui.cortexcn.dev/mcp"
    }
  }
}

Or add it from a terminal:

code --add-mcp "{\"name\":\"cortexcn\",\"type\":\"http\",\"url\":\"https://ui.cortexcn.dev/mcp\"}"

Windsurf and Devin Desktop

Add it to mcp_config.json:

{
  "mcpServers": {
    "cortexcn": {
      "serverUrl": "https://ui.cortexcn.dev/mcp"
    }
  }
}

OpenAI Codex CLI

codex mcp add cortexcn --url https://ui.cortexcn.dev/mcp

Or add it to ~/.codex/config.toml:

[mcp_servers.cortexcn]
url = "https://ui.cortexcn.dev/mcp"

Zed

Add it to your Zed settings.json:

{
  "context_servers": {
    "cortexcn": {
      "url": "https://ui.cortexcn.dev/mcp"
    }
  }
}

ChatGPT

Custom connectors need developer mode, which is available on the web for Plus, Pro, Business, Enterprise, and Edu plans.

1. Go to Settings > Security and login and turn on Developer mode.
2. Add a connector, paste https://ui.cortexcn.dev/mcp as the URL, and choose No authentication.

Example prompts

Once the server is connected, ask in plain words:

• "Find a hero block with a product screenshot and add it to the landing page."
• "Add the Lanyard 3D component and use our team's names."
• "Build a dashboard page with an area chart and three stat cards from Cortexcn."
• "What props does the Gooey Nav take?"
• "Show me the source of the Sphere Menu before we install it."
• "Which Cortexcn chart fits a sales funnel? Add it to the reports page."

Install from a component page

Every component page has an install box with an MCP tab. It shows the command that connects the server in Claude Code and a prompt to paste into your assistant, such as:

Add the Cortexcn lanyard-3d item to this project.

Assistants without MCP

If your assistant cannot use MCP, give it the docs as markdown instead:

• Add .md to any docs page URL, for example https://ui.cortexcn.dev/docs/components/lanyard-3d.md.
• Every docs page in one file: https://ui.cortexcn.dev/llms-full.txt.

Requirements

• An AI assistant that supports MCP.
• To install items: a React and Tailwind CSS project with shadcn initialized (npx shadcn@latest init).
• Node.js 18 or later for the shadcn CLI.

Related

• Skills: teach your assistant how to work with Cortexcn
• DESIGN.md: the design system your assistant should follow
• Installation: set up the @cortexcn registry by hand