UE Wingman

UE Wingman is a tool that allows an AI to control the unreal editor. When you're using it, it feels like the AI is right there looking at the editor with you. You'll be able to watch as it creates graph nodes and wires them together, you'll see it add components to your blueprints, you'll see it design widget hierarchies for you, and you'll see it write shaders for your materials.

The tool is not complete, not by a long shot. There are tons of Unreal Editor functions that the AI just can't access yet. Even so, I think it's useful: it has pretty comprehensive support to allow the AI agent to help create blueprints, widget blueprints, and materials.

How Does it Work?

This tool adds a command interpreter plugin to the Unreal Editor. You can type commands, and the plugin in the editor will execute them. You can actually type commands directly from the command-line. Here's an example of what you might see:

 $ ue-wingman.py Graph_Dump Graph=/Game/Testing/BP_Test,graph:EventGraph
 
 node K2Node_Event_0: Event BeginPlay
     output-pins OutputDelegate
 
 node K2Node_Event_2: Event Tick
    output-pins OutputDelegate, DeltaSeconds

There are tons of commands built in: Graph_Dump, GraphNode_Create, GraphPin_Connect, BlueprintComponent_Create, Widget_Create, and so forth. Using these commands, it's possible to examine and modify blueprints, widgets, and materials.

But, of course, these commands aren't really intended for humans. They're intended for an AI agent. The AI is given access to these commands using what's called "Model Context Protocol," which is a goofy name for "a mechanism that an AI can use to send commands to other software."

Why Choose this Particular Unreal Engine MCP?

There are a lot of Unreal Engine MCPs out there. Some of them are, shall we say, not carefully engineered. I'm a reasonably skilled software engineer and I've designed this plugin to be robust and capable of sustained development.

This MCP is also designed to be as broadly general as possible. I've seen many Unreal Engine MCPs that claim "can create 22 different kinds of graph nodes!" This makes me ask: why not just provide the entire catalog of all possible graph nodes? I've seen MCPs claim "you can edit 15 different material expression properties!" Why not provide access to all editable material expression properties? I've tried to make every tool in this MCP as capable as possible, with as few limits as possible.

This MCP is very extensible. Adding a new command requires a relatively small amount of code. I'm hoping some others in the community will eventually start contributing new commands.

Installation

There are three parts to UE Wingman:

The Unreal Plugin, which does 99% of the work.
The python program "ue-wingman.py" which a human can use to send commands to the plugin.
The python program "ue-wingman-mcp.py", which an AI can use to send commands to the plugin.

If you build Unreal from source, the best way to install the plugin is to drop the entire UEWingman source folder into your Plugins folder. Then do a build. Restart the editor, and go into your plugins configuration. Enable the UE Wingman plugin. You're done.

If you don't build from source, then unfortunately, you're out of luck. Precompiled plugins must be built for every different OS, for every different engine version. I just don't have the means to do that right now.

After installing the plugin, you need to install the two python programs. They are both short and simple: all they do is establish a network connection to the plugin, and then send the command you typed. They require python 3.6 or later, and no other dependencies.

To install the human version, ue-wingman.py, just drop it into a folder on your PATH.

To install the AI version, ue-wingman-mcp.py, you have to usually set up some config file for your AI agent. I use Claude Code, for that, you have to create a file ".mcp.json" in your project folder, and it needs to have this inside it:

{
  "mcpServers": {
    "ue-wingman": {
      "command": "python3",
      "args": ["Plugins/UEWingman/ue-wingman-mcp.py"]
    }
  }
}

You can usually ask your AI agent for help creating this config file.

The "User Manual"

You might be interested in seeing the "user manual" for the plugin. To get that, you type this:

$ ue-wingman.py UserManual

Of course, you're not the intended user: your AI agent is. When the AI agent starts up ue-wingman-mcp, it is automatically told to read the user manual. From there, the User Manual says, among other things, that the AI agent can get a listing of built-in commands. You can see that too:

$ ue-wingman.py ShowCommands

With these two commands at your disposal, you'll have a better understanding of what exactly your AI agent is doing with this plugin, and how it all works.

Fun things to Try

I really started enjoying this plugin when I asked my agent to make me a "cool looking material, something psychedelic and weird." It made a neat kaleidoscope thing. I then asked it to make me an animated rendering of the mandelbrot set. It's fun to watch it do things like that.

History and Credits

When I myself needed an MCP for unreal development, I did a survey of the plugins out there. I ended up choosing a plugin called "Blueprint MCP" by David Gundry:

https://github.com/mirno-ehf/ue5-mcp

It was not bad, but it had some limitations, and I started doing work to improve it. Incrementally, I ended rewriting pretty much the whole thing. So this whole project is actually a fork of Blueprint MCP. There's very little of the original code remaining. However, you will find snippets here and there.

Even though I ended up rewriting most of the code, it really was useful to have a functioning starting point. It meant I could improve one thing at a time, without having to try to get everything working all at once. So I'm quite grateful to David Gundry and his work.

Software License

UE Wingman is licensed under the MIT license, a copy of which is enclosed. Its predecessor (of which it is a fork) was also under the MIT license, so everything works out.

6.1 KiB Raw Blame History