🔌 Setting Up Unreal MCP

Before You Begin

Setting up Unreal MCP has two halves. Inside Unreal, you enable a plugin and let it run a small server. Outside Unreal, you point an AI client (such as Claude Code, Cursor, or VS Code) at that server using a configuration file. The good news is that Unreal can write most of that configuration for you.

Here is the whole process at a glance before we walk through each step in detail:

flowchart LR
    A["1. Enable the
Unreal MCP plugin"] --> B["2. Configure the
server settings"]
    B --> C["3. Generate the
client config"]
    C --> D["4. Verify the
live bridge"]
    D --> E["✅ Ready for
AI-assisted work"]

    style A fill:#667eea,color:#fff
    style E fill:#4CAF50,color:#fff
                

Figure: The four setup steps in this lesson. Each one has a real screenshot so you can confirm your editor matches.

Step 1: Enable the Unreal MCP Plugin

Open the Plugins browser from the main menu: Edit · Plugins. This window lists every plugin the editor can see. Type Unreal MCP into the search box in the top-right to filter the list down to the one you want.

You are looking for the plugin named Unreal MCP, described as "Anthropic MCP (Model Context Protocol) server implementation for Unreal Engine" and published by Epic Games, Inc. Tick its checkbox to enable it.

Figure: The Unreal MCP plugin enabled in the Plugins browser. Note the purple Experimental badge next to its name. · Experimental in UE 5.8.

Step 2: Configure the Server

With the plugin enabled and the editor restarted, its settings live in Editor Preferences. Open Edit · Editor Preferences, then select Model Context Protocol from the list on the left. (If you cannot find it, type "Model Context" into the settings search box at the top.)

The settings are grouped into Server and Tools:

Figure: The Model Context Protocol settings in Editor Preferences. The defaults are all you need to get started. · Experimental in UE 5.8.

Server Url Path is the path portion of the address, /mcp by default. Server Port Number is 8000 by default. Together with the localhost host, these form the full address a client connects to:

http://127.0.0.1:8000/mcp

Auto Start Server is on by default, which means the embedded server starts automatically each time you open the editor. That is exactly what you want for everyday use, so leave it checked. Enable Tool Search (under Tools) lets clients discover tools through the Toolset Registry, as described in Lesson 11.1.

Step 3: Generate the Client Configuration

Your AI client needs to know where the server is. Rather than writing that configuration by hand, the plugin ships a console command that generates it for you. Open the editor console (the Cmd box at the bottom of the Output Log, or the command bar in the status bar) and run:

ModelContextProtocol.GenerateClientConfig ClaudeCode

The argument names which client you are configuring. The supported targets are ClaudeCode, Cursor, VSCode, Gemini, Codex, and All. Use the one that matches the assistant you plan to connect, or All to write configuration for every supported client at once.

The command writes an .mcp.json file into your project folder and reports the location in the Output Log:

Figure: Running the GenerateClientConfig command writes an .mcp.json client configuration into the project folder and logs where it went. · Experimental in UE 5.8.

That .mcp.json file is a small text file describing the server address (http://127.0.0.1:8000/mcp) in the format your client expects. Point your client at it, or place your client's config where it looks for one, and the client will know how to reach the editor.

Step 4: Verify the Live Bridge

Before you rely on the bridge, confirm it is actually running. The clearest proof lives in the Output Log. Open it (the Output Log button in the status bar, or Window · Output Log) and type ModelContextProtocol into its Search Log box to filter out everything else.

When a client connects and starts working, the server logs each step: it negotiates a protocol version, initializes a session, and dispatches each tool call to the right toolset. Seeing these lines is your confirmation that the bridge is live end to end.

Figure: The Output Log filtered to ModelContextProtocol, showing the live server negotiate a session and dispatch toolset tools. Each "Dispatching toolset tool" line is one action the agent asked the editor to perform. · Experimental in UE 5.8.

Read a few of those lines and the architecture from Lesson 11.1 comes to life. Client requested protocol version, negotiated is the handshake. Session initialized means a client is connected. Each Dispatching toolset tool line names exactly which toolset and tool ran, for example a logs tool or a UI-inspection tool. That is the Toolset Registry routing a request to the right place.

Troubleshooting

If something is not working, most problems fall into a handful of categories. Work through these in order.

The plugin is not in the list

Clear the Plugins search box and confirm you are searching by the exact name, Unreal MCP. Because it is an Experimental, Built-In plugin, make sure the browser is not hiding built-in plugins through its filters.

Settings or server missing after enabling

Enabling a plugin only takes effect after an editor restart. If the Model Context Protocol settings do not appear in Editor Preferences, or the server never starts, restart the editor and try again.

The server does not start

Check that Auto Start Server is enabled in Editor Preferences (Step 2). Then open the Output Log filtered to ModelContextProtocol and look for warnings or errors. A common cause is a port already in use: if port 8000 is taken, change the Server Port Number and regenerate the client configuration.

The client cannot connect

Confirm the address is exactly http://127.0.0.1:8000/mcp (or your custom port). Regenerate the configuration with GenerateClientConfig so the client picks up the current settings. Remember the server is localhost-only by default, so the client must run on the same machine as the editor.

Knowledge Check

Summary

You have turned the concept from Lesson 11.1 into a working setup. Here is the path you followed:

Enable the plugin. In Edit · Plugins, search for and enable Unreal MCP (Experimental), then restart the editor so the change takes effect.

Configure the server. In Edit · Editor Preferences · Model Context Protocol, the server defaults to /mcp on port 8000 with Auto Start Server enabled, giving the address http://127.0.0.1:8000/mcp.

Generate the client configuration. Run ModelContextProtocol.GenerateClientConfig with your client name to write an .mcp.json into the project folder.

Verify the live bridge. Filter the Output Log to ModelContextProtocol and watch the server negotiate sessions and dispatch toolset tools. Those lines are proof the bridge is running.

What's Next?

The bridge is built and verified. In the final lesson of this module, Your First AI-Driven Workflow, you will put it to work: give an assistant a plain-language goal and watch it author something real in your project through the very toolsets you just saw dispatching in the log.