# What is a design MCP?

*How-to · by The Brilliant team · updated Nov 20, 2026 (2026-11-20)*

> **Short answer.** A design MCP is a Model Context Protocol server that gives an AI agent, such as Claude Code, Cursor or Codex, tools to read and change a design inside a design tool, so the agent's work lands as layers on a canvas and not only as code. Brilliant is one: its MCP server lets an agent create, edit, read and export real vector designs that you can then change by hand.

## How to try a design MCP in four steps

### 1. Connect the server to your agent

For Claude Code, run the line below in a terminal. It needs Node 20 or later and no API key. Every other client has its own line on the Supported clients page.

```text title="Terminal"
claude mcp add --scope user brilliant -- npx -y @brilliant-hq/mcp
```

### 2. Ask for a design

Name the tool in the request, so the agent designs on the canvas and does not write the screen as code.

```text title="Prompt"
Design a three-tier pricing section in Brilliant.
```

### 3. Watch the tool calls

The agent calls the server's tools: init to start a session, edit to write the design, read to look at the canvas and export to get PNG, SVG, PDF, HTML or React. The first call opens brilliant.design in Chrome, Edge or Firefox and pairs with it, or uses the desktop app when it is running.

### 4. Edit by hand, then hand it back

What the agent made is frames, auto layout, components and text you can drag, retype and restyle. The agent reads your changes and keeps going from them.

## Common problems

| Problem | Fix |
|---|---|
| The server only reads the design | Some design MCP servers give an agent the design as context for writing code and stop there. Check for write tools before you plan on the agent designing. |
| The free plan runs out mid-task | Vendors meter agent work in their own units: tool calls a month, tool calls a week, agent actions a day or days a month. Read each vendor's line on the MCP limits page before you choose. |
| The agent writes code instead of designing | Name the design tool in the prompt, for example: Design it in Brilliant. |

## Questions

### What does MCP stand for?

Model Context Protocol, an open protocol that lets an AI application call tools on a server. A design MCP is a server whose tools belong to a design tool.

### Is a design MCP the same as a design tool plugin?

No. A plugin runs inside the design tool's window when you launch it. A design MCP server is called by your agent from wherever the agent runs, such as a terminal or a code editor.

### Can a design MCP write to the canvas, or only read it?

It depends on the server. On November 20, 2026, Figma, Penpot, Paper, Pencil, Subframe, Kite, Framer and Brilliant each documented tools that write to the canvas. Figma says its canvas writes are free during a beta and will become a usage-based paid feature.

### Is the Brilliant MCP server free?

Yes. It is free on every plan, with unlimited canvas work by your own agents. An image an agent generates on Brilliant AI uses your Brilliant AI usage.

### Which agents can use a design MCP?

Any client that speaks MCP. Brilliant has a connect line for Claude Code, Cursor, VS Code, Codex, Gemini CLI, Antigravity, Windsurf, Zed, OpenCode, Amp and Claude Desktop, and it works in ChatGPT and Claude with no local setup.

## Related

- [MCP servers for design work in 2026](https://brilliant.design/blog/mcp-servers-for-design)
- [MCP limits: what is free and what is capped](https://brilliant.design/docs/agents/limits)
- [Supported clients, with each connect line](https://brilliant.design/docs/agents/clients)
- [MCP design tool](https://brilliant.design/mcp)
- [How to make Claude Code design a UI](https://brilliant.design/learn/design-a-mobile-app-screen-with-claude-code)

[Open the editor](https://brilliant.design/playground)
