Unity AI Context Bridge MCP

Node MCP server that feeds AI tools Unity GameObjects, assets, and other resources that are selected by the user. Think of it like when you drag source code in Cursor, Antigravity, Github Copilot window to add that file to context, but for Unity.

[!IMPORTANT]
The MCP server requires the Unity AI Context Bridge Unity package to be installed in the Unity project. See more details at https://github.com/jamius19/unity-ai-context-bridge

[!TIP]
For the best experience, use Unity Official MCP server along with this.

NPM Package Page at https://npmjs.com/package/@jamius19/unity-ai-context-bridge-mcp

Installation

Install the package globally:

npm install -g @jamius19/unity-ai-context-bridge-mcp

Then the MCP server will be available on your terminal with the uacb-mcp command.

MCP Client Installation

Install the NPM package globally before wiring an MCP client to it. This will enable the uacb-mcp command in your terminal,

Choose one transport per client entry:

1. Stdio: the MCP client launches this server as a subprocess. (Recommended)
2. Streamable HTTP: you start this server with uacb-mcp --http, then the MCP client connects to http://127.0.0.1:3333/mcp.

GitHub Copilot (VS Code)

Create or update .vscode/mcp.json in a project, or run MCP: Open User Configuration from the Command Palette for a user-wide setup.

Stdio:

{
  "servers": {
    "unity-ai-context-bridge": {
      "type": "stdio",
      "command": "uacb-mcp",
      "args": ["--stdio"]
    }
  }
}

HTTP with auth:

{
  "servers": {
    "unity-ai-context-bridge": {
      "type": "http",
      "url": "http://127.0.0.1:3333/mcp",
      "headers": {
        "Authorization": "Bearer change-me"
      }
    }
  }
}

HTTP without auth:

{
  "servers": {
    "unity-ai-context-bridge": {
      "type": "http",
      "url": "http://127.0.0.1:3333/mcp"
    }
  }
}

After adding the server, run MCP: List Servers from the Command Palette and start unity-ai-context-bridge if VS Code does not start it automatically. Then open GitHub Copilot Chat in Agent mode and enable the server tools from the tools menu if needed.

Rider/Other mcp.json clients

Many MCP clients, including Claude Desktop-style JSON configs, accept the same mcpServers shape.

Stdio:

{
  "mcpServers": {
    "unity-ai-context-bridge": {
      "command": "uacb-mcp",
      "args": ["--stdio"]
    }
  }
}

HTTP:

{
  "mcpServers": {
    "unity-ai-context-bridge": {
      "url": "http://127.0.0.1:3333/mcp",
      "headers": {
        "Authorization": "Bearer change-me"
      }
    }
  }
}

Claude Code

Stdio:

claude mcp add --transport stdio unity-ai-context-bridge -- uacb-mcp --stdio

HTTP with auth:

claude mcp add --transport http --header "Authorization: Bearer change-me" unity-ai-context-bridge http://127.0.0.1:3333/mcp

HTTP without auth:

claude mcp add --transport http unity-ai-context-bridge http://127.0.0.1:3333/mcp

Codex

Add one of these entries to ~/.codex/config.toml.

Stdio:

[mcp_servers.unity_ai_context_bridge]
command = "uacb-mcp"
args = ["--stdio"]
enabled = true

HTTP with auth:

[mcp_servers.unity_ai_context_bridge]
url = "http://127.0.0.1:3333/mcp"
http_headers = { Authorization = "Bearer change-me" }
enabled = true

HTTP without auth:

[mcp_servers.unity_ai_context_bridge]
url = "http://127.0.0.1:3333/mcp"
enabled = true

Gemini CLI

Stdio:

gemini mcp add unity-ai-context-bridge uacb-mcp --stdio

HTTP with auth:

gemini mcp add --transport http unity-ai-context-bridge http://127.0.0.1:3333/mcp --header "Authorization: Bearer change-me"

HTTP without auth:

gemini mcp add --transport http unity-ai-context-bridge http://127.0.0.1:3333/mcp

Cursor

Create or update .cursor/mcp.json in a project, or ~/.cursor/mcp.json for a user-wide setup.

Stdio:

{
  "mcpServers": {
    "unity-ai-context-bridge": {
      "type": "stdio",
      "command": "uacb-mcp",
      "args": ["--stdio"]
    }
  }
}

HTTP with auth:

{
  "mcpServers": {
    "unity-ai-context-bridge": {
      "url": "http://127.0.0.1:3333/mcp",
      "headers": {
        "Authorization": "Bearer change-me"
      }
    }
  }
}

HTTP without auth:

{
  "mcpServers": {
    "unity-ai-context-bridge": {
      "url": "http://127.0.0.1:3333/mcp"
    }
  }
}

Run over stdio

uacb-mcp --stdio

Use this transport from MCP clients that launch servers as subprocesses.

Run over HTTP

In HTTP mode it defaults to http://127.0.0.1:3333/mcp URL.
You can customize it via the following ways,

PowerShell:

$env:MCP_AUTH_TOKEN = "change-me"
$env:PORT = "3333"
uacb-mcp --http

Bash:

MCP_AUTH_TOKEN=change-me PORT=3333 uacb-mcp --http

If you do not want HTTP auth during local development, omit MCP_AUTH_TOKEN:

uacb-mcp --http

HTTP endpoints:

• GET /health does not require auth.
• GET /context returns the raw currently selected Unity objects/assets context JSON. It requires a projectPath query parameter to select a specific Unity project bridge.
• POST /mcp is the SDK Streamable HTTP MCP endpoint for AI tools.

Authenticated HTTP requests must include either:

Authorization: Bearer change-me

or:

x-auth-token: change-me

MCP surface

Resource template:

• unity-context://projects/items{?projectPath}: currently selected Unity GameObjects/assets of interest for a specific project path. Listed concrete resources percent-encode the project path in the query string, so paths with spaces are represented safely, for example unity-context://projects/items?projectPath=C%3A%5Cexample%5Cunity%5Cunity%5Cproject for an Unity project in the C:\example\unity\unity\project path.

Tool:

• get_unity_context_items: returns the current Unity selection context for AI tools. It requires a URL-encoded projectPath; the server reads that project's bridge file and uses its URL.

Example JSON-RPC request:

{
  "jsonrpc": "2.0",
  "id": 1,
  "method": "tools/call",
  "params": {
    "name": "get_unity_context_items",
    "arguments": {
      "projectPath": "C%3A%5Cexample%5Cunity%5Cunity%5Cproject"
    }
  }
}

If you have Unity Official MCP installed, this will be done automatically, otheriwse you can update your AGENTS.md file with the full project path.

How does it work?

It automatically discovers the live Unity selection context from the bridge file under the requested Unity project path:

<projectPath>/Temp/unity-ai-context-bridge/bridge.json

The bridge file contains the local URL and auth key for that Unity editor instance. The MCP server reads that file for the requested projectPath, connects to its url, and sends the auth key to Unity AI Context Bridge.

This bridge file is auto generated by the Unity AI Context Bridge Unity Editor Extension.