Funplay Unreal MCP server

Funplay MCP for Unreal The most advanced MCP server for the Unreal Editor. 中文 | English

💖 If this saves you time, please star the repo — it helps a lot.

Funplay MCP for Unreal lets AI assistants — Claude Code, Cursor, Windsurf, Codex, VS Code Copilot and any other MCP client — operate directly inside a running Unreal Editor. It is a pure-Python editor plugin (no C++ build, works across Unreal Engine 5.3–5.8) that exposes the editor over a local MCP server, plus a tiny Node.js stdio bridge that connects your AI client to it. MIT-licensed and free for commercial use.

Describe what you want in natural language — spawn and arrange actors, build levels and Blueprints, create materials, import assets, drive Play-In-Editor, capture screenshots, or run arbitrary executepython* — and let the assistant do it.

"Build a small arena: a floor, four walls, some point lights, and a player

start in the middle. Then play it and send me a screenshot."

The assistant inspects the level, spawns and positions the actors, sets up

lighting, enters Play-In-Editor, and returns a screenshot — all through this

MCP server.

Demo

Codex asks for a castle, Funplay MCP drives the Unreal Editor through MCP, and the finished scene appears directly in the editor.

Quick Start

Three things: (1) drop the plugin into your project, (2) enable it and start the server, (3) point your AI client at it.

1. Install the plugin

💡 The plugin is pure Python — there is nothing to compile.

(/Plugins/FunplayMCP/), or download Funplay.UnrealMcp.vX.Y.Z.zip from the latest Release and extract it there.

• Copy the FunplayMCP/ folder into your project's Plugins/ directory

.uplugin enables automatically.

• It depends on the engine's Python Editor Script Plugin, which the

2. Enable and start the MCP Server

enabled, then restart the editor if prompted.

• Open your project. In Edit → Plugins, confirm Funplay MCP for Unreal is

port is busy it automatically picks a free local port (and a second editor of the same project attaches to the existing server).

• On startup the server listens at http://127.0.0.1:8765/ by default. If the

/Saved/FunplayMCP/funplaymcpsettings.json.

• A per-project auth token is generated and stored in

• Get the endpoint and token any time from **Tools → Funplay MCP → Log Endpoint
• Token** (printed to the Output Log).

3. Configure your AI client

The fastest path is the editor menu: Tools → Funplay MCP → Configure Claude Code (or Cursor / VS Code / Codex). It writes the right config with the endpoint and token filled in.

Manual configuration

Claude Code (~/.claude.json), Cursor (~/.cursor/mcp.json):

{
  "mcpServers": {
    "funplay-unreal": {
      "command": "npx",
      "args": ["-y", "funplay-unreal-mcp"],
      "env": {
        "FUNPLAY_UNREAL_MCP_URL": "http://127.0.0.1:8765/",
        "FUNPLAY_UNREAL_MCP_TOKEN": "<token from the Funplay MCP menu>"
      }
    }
  }
}

Shortened. The full README is on GitHub.