UnityFigmaMCP

Build Unity UI prefabs from Figma designs using AI agents via Model Context Protocol.

An AI agent reads your Figma file, talks to the Unity Editor in real time, and assembles
prefabs — with layout, text, images, components, and variants — without you writing a
single line of code.

┌─────────────┐      MCP (stdio)      ┌─────────────────┐    SignalR     ┌──────────────┐
│  AI Agent   │ ◄───────────────────► │   MCP Server    │ ◄────────────► │ Unity Editor │
│             │                       │    (.NET 8)     │   localhost    │   Package    │
└─────────────┘                       └────────┬────────┘                └──────────────┘
                                               │
                                        Figma REST API
                                               │
                                      ┌────────▼────────┐
                                      │   Figma File    │
                                      └─────────────────┘

How it works

1. You give the agent a Figma URL and say "build a prefab"
2. The MCP server fetches the design tree from Figma
3. It sends the layout data to Unity Editor via SignalR
4. Unity runs the pipeline — RectTransform, Text, Image, Layout Groups, ContentSizeFitter — and saves a .prefab
5. Component instances are resolved automatically: if a sprite or prefab is already bound, it gets reused

The whole thing runs locally. Your Figma token stays on your machine.

Features

13 MCP tools for full Figma-to-Unity control:

6 AI skills — ready-made workflows for any MCP-compatible agent:

Layout pipeline with configurable steps:
RectTransform · Text (TextMeshPro) · Image · Horizontal/Vertical Layout Groups · Grid · ContentSizeFitter

Quick Start

Prerequisites

• Unity 2021.3 or later
• .NET 8 SDK
• An MCP-compatible AI agent (Claude Code, Cursor, Windsurf, or any other)
• Figma access token

1. Install the MCP Server

Clone this repository outside your Unity project (it's not a Unity asset — it's a standalone server):

git clone https://github.com/chav0/UnityFigmaMCP.git

Add the server to your MCP client config.

Claude Code (~/.claude.json or project .mcp.json):

{
  "mcpServers": {
    "unity-figma-mcp": {
      "command": "/path/to/UnityFigmaMCP/UnityFigmaMCPServer/run-mcp.sh",
      "env": {
        "FIGMA_ACCESS_TOKEN": "your-figma-token"
      }
    }
  }
}

On Windows, use run-mcp.bat instead of run-mcp.sh.

2. Install Dependencies

Open your Unity project, then install these packages via Window > Package Manager > + > Add package from git URL:

1. NuGetForUnity — NuGet package manager for Unity:

   https://github.com/GlitchEnzo/NuGetForUnity.git?path=/src/NuGetForUnity
   

2. SerializeReference Extensions — required for pipeline step selection in the Inspector:

   https://github.com/mackysoft/Unity-SerializeReferenceExtensions.git?path=Assets/MackySoft/MackySoft.SerializeReferenceExtensions
   

3. SignalR Client — open NuGet > Manage NuGet Packages, search for Microsoft.AspNetCore.SignalR.Client and install it

3. Install the Unity Package

In Package Manager, add one more git URL:

https://github.com/chav0/UnityFigmaMCP.git?path=UnityFigmaMCPPackage

To pin a specific version:

https://github.com/chav0/UnityFigmaMCP.git?path=UnityFigmaMCPPackage#v0.1.1

The package auto-connects to the MCP server on port 52802 when the Editor starts.

Note: The cloned repo and the Unity package are separate things. The repo contains the MCP server source code; the Unity package is installed independently via UPM from the same git URL but a different path. They don't conflict.

4. Configure a Layout Pipeline

Open the FigmaAutoLayoutSettings asset (Assets/UnityFigmaMCP/Editor/) in the Inspector
and add at least one pipeline. A basic pipeline for screens includes these steps:

Text · RectTransform · Image · Vertical Layout Group · Horizontal Layout Group · Grid

Also set Prefab Folder and Sprites Folder to where you want generated assets saved
(e.g. Assets/Prefabs and Assets/Sprites).

Without a pipeline, unity_build_prefab has nothing to run and the build will fail.

5. Bind Your UI Kit

Before building prefabs, bind your existing sprites and prefabs to Figma components.
This lets the build pipeline reuse your assets instead of generating placeholders.

Use the /bind-figma-unity skill, or call unity_bind manually:

Bind my UI kit sprites and prefabs to Figma components from file abc123, node 100:200

This step is important — without binding, the agent won't know which Unity assets
correspond to which Figma components, and builds will have missing references.

6. Add Skills (Optional)

Skills are Markdown workflow files that teach the agent to use the MCP tools effectively.
They work with any agent that supports skill/instruction files — not just Claude Code.

Claude Code — copy into your project:

cp -r /path/to/UnityFigmaMCP/.claude/skills/* /path/to/your-project/.claude/skills/

Other agents — copy the skill Markdown files into your agent's instruction/prompt directory.
The skills are plain Markdown with no agent-specific dependencies.

7. Verify

Open Unity, then ask your agent:

Check the project status

This runs the status tool and confirms both the Figma token and Unity connection are working.

Documentation

• Installation Guide — detailed setup for macOS and Windows
• Tools Reference — all 13 MCP tools with parameters and examples
• Skills Guide — how to use the 6 AI skills
• Architecture — how the three components connect

Updating

Unity Package — change the tag in your Packages/manifest.json:

"com.hugglebit.unity.figma.mcp": "https://github.com/chav0/UnityFigmaMCP.git?path=UnityFigmaMCPPackage#v0.1.1"

MCP Server — pull and rebuild:

cd UnityFigmaMCP
git pull

The run-mcp.sh script rebuilds automatically on each launch.

Skills — re-copy the skill files from the updated repo.

Contact

Questions, ideas or bug reports are welcome.

• Email — [email protected]
• Telegram — @alinulken
• LinkedIn — achupakhina

License

MIT