AI & IT toolsAI Tools

Letting Claude Code work in Unreal Engine 5.8 with the MCP plugin: from enabling the experimental feature to placing three cubes

Enable the Unreal MCP plugin that arrived in Unreal Engine 5.8 as an experimental feature, connect Claude Code, get a summary of the level, and have it place three cubes. It also covers what to watch out for with an experimental feature and how to undo changes.

AZTechnology13 min read

한국어English

Based on: Unreal Engine 5.8 (screenshots, Korean interface), the Unreal MCP plugin 1.0 (experimental), Claude Code 2.1.316, and Epic's official page "Unreal MCP in Unreal Editor". Checked in November 2026. All pictures are real screens from a laptop with an 8 GB graphics card. The red numbers and boxes were added to explain them. The editor in the pictures shows the Korean interface, so the English menu names are given in the text.

Unreal Engine 5.8 ships with an official plugin that lets AI connect straight to the editor. The release notes heading is "MCP Server (Experimental)", and the plugin's display name is Unreal MCP. The release notes say it "enables agentic AI systems to connect to the Unreal Editor".

MCP (Model Context Protocol) is a common standard for plugging tools into AI. Just as one plug shape lets any appliance use the same socket, any AI app that supports MCP connects to Unreal the same way. The full idea is in What is MCP?. If you have read this site's Blender MCP guide (Connecting the official Blender MCP to Claude Code), this is the same idea moved to Unreal.

This plugin is an experimental feature. The official page says to use caution when shipping with it, and notes that the server has no authentication layer, so it is not safe to expose it beyond your own computer. This guide therefore assumes you only follow along in a practice copy of a project.

By the end of this guide you will be able to:

  1. Enable the Unreal MCP and All Toolsets plugins.
  2. Start the MCP server automatically and check its address and port.
  3. Create the settings file Claude Code reads (.mcp.json).
  4. Get a level summary from Claude Code and have it place cubes.
  5. Undo what the AI changed and use the experimental feature safely.

1. How the pieces fit

PieceWhat it doesRestaurant comparison
Claude CodeUnderstands your request and decides which tool to useThe waiter taking orders
Unreal MCP server (inside the editor)Turns the AI's tool requests into editor actionsThe kitchen counter
Toolsets (All Toolsets)The actual features, such as placing actors or moving the cameraThe cooking tools
Unreal EditorActually changes the levelThe kitchen

According to the official page, Unreal MCP itself is only the server, and the actual tools come from toolset plugins. That is why you enable both plugins. The server opens only inside your computer at http://127.0.0.1:8000/mcp. 127.0.0.1 is the address that means "this computer itself".

2. Preparation: a practice project

Starting point: Unreal Engine 5.8 installed from the Epic Games Launcher.

Steps:

  1. Start Unreal Engine 5.8 from the launcher and create a new project. Give it a name that marks it as practice, such as UEDemo.
  2. If you want to test with a project you are already working on, copy the whole project folder first and use the copy.

The pictures in this guide were taken on the default open world level, with wide terrain already in place. Depending on the template and level you pick, the number of actors on your screen may differ from the pictures.

On a laptop with an 8 GB graphics card, close other heavy programs (Blender, games, and so on) before you start. On my PC, one editor with the default level open used about 4.2 GB of memory.

3. Enable two plugins

Starting point: the practice project open in the editor.

Steps:

  1. Choose Edit ‣ Plugins from the top menu (② in Figure 1).
  2. Type Unreal MCP into the search field of the Plugins window. Tick the checkbox on the Unreal MCP row (① in Figure 2). An Experimental tag sits next to the name (② in Figure 2).
  3. Clear the search field, type All Toolsets, and tick the All Toolsets row too (Figure 3).
  4. Restart the editor when asked.

The Edit menu (Korean interface). The History group shows Undo object placement CTRL+Z, Redo, and Undo History; the Configuration group shows Editor Preferences CTRL+ALT+E, Keyboard Shortcuts, Project Settings, and Plugins Figure 1. The Edit menu (Korean interface). ① Editor Preferences ② Plugins. The picture was taken after the AI placed the cubes, so an undo entry shows at the top (Unreal Editor screenshot)

The Plugins window (Korean interface). Unreal MCP in the search field; one result row with a ticked checkbox, an icon, the name Unreal MCP with a purple Experimental tag, the description Anthropic MCP (Model Context Protocol) server implementation for Unreal Engine, and Version 1.0 Epic Games, Inc. on the right Figure 2. The Unreal MCP plugin (Korean interface). ① Enable checkbox ② Name and Experimental tag (Unreal Editor screenshot)

Plugins window search results (Korean interface). All Toolsets in the search field; the first row All Toolsets (Experimental, description Aggregator plugin that depends on all Toolsets plugins.), the second row Niagara Toolsets Figure 3. The All Toolsets plugin (Korean interface). ① Enable checkbox ② Name. Enabling All Toolsets turns on the toolset plugins it depends on (Unreal Editor screenshot)

All Toolsets turns on many toolset plugins at once. The official page says you can also enable only the toolsets you need. At first it is easier to understand if you turn them all on.

4. Start the MCP server

Starting point: the editor restarted with both plugins enabled.

Steps:

  1. Open Edit ‣ Editor Preferences (① in Figure 1).
  2. In the list on the left, choose Model Context Protocol under General (① in Figure 4).
  3. Check that Server URL Path is /mcp and Server Port Number is 8000 (② in Figure 4).
  4. Turn on Auto Start Server (③ in Figure 4). From now on the server starts every time the editor opens.
  5. To start it right now, type ModelContextProtocol.StartServer into the console field at the bottom of the editor (default key `) and press Enter.

The Editor Preferences window (Korean interface). Model Context Protocol is selected in the list on the left; on the right, the Server group shows Server URL Path /mcp, Server Port Number 8000, and an Auto Start Server checkbox (off), and the Tools group shows Enable Tool Search (on) Figure 4. MCP server settings (Korean interface). ① Model Context Protocol ② Path /mcp and port 8000 ③ Auto Start Server. The picture shows the screen before auto start was turned on (Unreal Editor screenshot)

When the server starts, the Output Log shows this line:

LogModelContextProtocol: Starting MCP server on port 8000 (override with -ModelContextProtocolPort=N).

If another program already uses port 8000, change the port number in Figure 4. Leave Enable Tool Search below it turned on. With it on, the server does not show hundreds of tools at once. It offers three tools, list_toolsets, describe_toolset, and call_tool, and finds what it needs through them. It is like reading the index of a big toolbox instead of dumping everything on the floor, so the AI has less text to read.

5. Create the Claude Code settings file

Starting point: the editor with the server running.

Steps:

  1. Type the line below into the console field and press Enter. According to the official page, this writes a .mcp.json file in the project folder (where the .uproject file is).
ModelContextProtocol.GenerateClientConfig ClaudeCode
  1. Open .mcp.json in the project folder with Notepad and check that it looks like this.
{ "mcpServers": { "unreal-mcp": { "type": "http", "url": "http://127.0.0.1:8000/mcp" } } }

Put Cursor, VSCode, Gemini, or Codex in place of ClaudeCode to write a settings file for another AI app, or All to write them all. A file you write by hand works the same if the content is the same. This server supports HTTP only (including SSE), not the command-line (stdio) connection method or WebSocket. SSE (Server-Sent Events) is an HTTP method where the server keeps the connection open and keeps sending updates.

If you have not installed Claude Code yet, see Claude Code basics first. You can also register the server with one command: run claude mcp add --transport http unreal-mcp http://127.0.0.1:8000/mcp in the project folder.

6. First request: a level summary

Starting point: the editor running, in the project folder that holds .mcp.json.

Steps:

  1. Make sure the editor is running first. The official page says that if the connection fails, start the AI app after the editor.
  2. Open a terminal, move to the project folder, and run claude. If it asks whether to use this folder's MCP server, allow it.
  3. Ask as below. For the first request, add do not change anything, so it only reads.
Using the unreal-mcp tools connected to the Unreal Editor, briefly summarize what kinds of actors are in the open level and how many of each. Don't change anything. (translated from Korean)

On my PC, the start of the answer Claude Code gave was:

The open level is a new, unsaved level (/Temp/Untitled_1), and 145 actors are loaded in memory right now.
I did not change the level; I only looked it up.
- LandscapeStreamingProxy (landscape pieces, 8×8): 64
- HLOD0_Instancing (automatically created HLOD actors): 64
- Lighting and environment: DirectionalLight, SkyLight, SkyAtmosphere, ExponentialHeightFog, VolumetricCloud, 1 each
- Actors the engine places automatically (WorldSettings, Brush, and so on): 9
Summary: it is the default open world template as is.
(translated from Korean)

The Outliner showed 138 (Figure 5), but the AI answered 145. The AI appears to have also counted engine actors that the Outliner does not show clearly, such as WorldSettings. When the AI's number and the screen's number differ, ask again what it counted.

The default level in the Unreal Engine 5.8 editor (Korean interface). Wide sand-colored terrain in the viewport; the Outliner on the right shows a Lighting folder, Landscape, PlayerStart, and 138 actors (138 loaded) Figure 5. The level before giving the AI any work (Korean interface) (Unreal Editor screenshot)

7. Second request: place three cubes

Starting point: the Claude Code screen after the level summary.

Steps:

  1. Ask as below. Say what to place, where, how many, what to name them, and not to save.
Using the unreal-mcp tools, place three engine basic cubes (/Engine/BasicShapes/Cube) in a row 1 m apart on the ground in the middle of what the current level camera sees, name them AI_Cube_1 to 3, then move the viewport camera so all three cubes are clearly visible. Don't save the level. When you're done, tell me what you did in three lines. (translated from Korean)
  1. Look at the editor. Check that three cubes were placed, the camera moved, and AI_Cube_1 to 3 appear in the Outliner (Figure 6).

Unreal Engine 5.8 viewport (Korean interface). Three white cubes in a row on a wide floor with long shadows; the Outliner on the right shows AI_Cube_1, AI_Cube_2, AI_Cube_3 and 141 actors (141 loaded); the tab at the top reads Untitled* Figure 6. After the AI's changes (Korean interface). ① The three placed cubes ② AI_Cube_1 to 3 in the Outliner (Unreal Editor screenshot)

The answer on my PC said it placed the three cubes where the middle of the original camera view met the ground, left 1 m of space between the cubes (2 m between centers), and asked me to say so if I had meant the cubes to touch. It read "1 m apart" as the gap between cubes. The basic cube is 1 m on each side, so the centers ended up 2 m apart. When a request can be read two ways, spell it out, for example "1 m between centers".

8. Undo and safety checks

The cubes the AI placed go into the undo history just like your own edits. "Undo object placement" at the top of the Edit menu in Figure 1 is that record. If you do not like the result, press Ctrl+Z.

CheckWhy
Work in a practice copy of the projectIt is experimental and may behave unexpectedly
Put "don't save" in requests, and save yourself after checkingSo a person reviews the changed level before keeping it
Never open the server port beyond your computerThe server has no authentication (official page)
Make the first request read-onlyTo see what the AI sees first
Turn the server off when you are not using itConsole: ModelContextProtocol.StopServer

General safety habits for handing a 3D program to AI are covered in more detail in Safety checks before letting AI work in Blender. They apply to Unreal as well.

When you get stuck

SymptomCauseFix
Claude Code cannot find the unreal-mcp toolsClaude Code started before the editor, or from another folderStart the editor first, then run claude again in the folder with .mcp.json
No Starting MCP server line in the Output LogAuto start is offTurn on Auto Start Server as in section 4, or use StartServer in the console
Tools show up but can do almost nothingAll Toolsets (or the toolsets you need) is not enabledEnable All Toolsets as in section 3 and restart the editor
The server does not start and reports a port errorAnother program uses port 8000Change the port number in Figure 4 and update the address in .mcp.json to match
The editor becomes very slowThe heavy default level and other programs share memoryClose other programs and test in a light empty level
The AI saved the levelThe request did not say not to saveGo back to the copy of the project, start over, and add "don't save" from the next request on

Summary

Enable Unreal MCP + All Toolsets, turn on Auto Start Server (127.0.0.1:8000/mcp), run GenerateClientConfig ClaudeCode in the console, start the editor first and run claude in the project folder, and only work in a copy, with a person doing the saving.

References

KEEP EXPLORING

Read next

Top of the Artificial Analysis text-to-image leaderboard AA-Image-T2I v2.1. Rank 1 GPT Image 2.5 Sunburst (max) 1222, rank 2 GPT Image 2.5 Flare (max) 1207, rank 3 GPT Image 2 (high) 1186, rank 4 Nano Banana 2.1 1176, rank 5 MAI-Image-2.6 1166, rank 6 Grok Imagine Image 2.0 1166, rank 7 Nano Banana 2 1140, price per 1,000 images on the rightAI & IT tools
AI Tools13 min read

Choosing an AI image model in autumn 2026: GPT Image 2.5 and Nano Banana 2.1 compared

Names, release dates, prices, and leaderboard positions of GPT Image 2.5 (Flare and Sunburst) and Nano Banana 2.1, the image models that arrived in autumn 2026, checked against official sources, plus simple rules for picking one by use.

A low-poly desert at sunset. Four launchers fire missiles, a three-ring aim circle with 90% sits among drones marked by red diamonds, and the top left shows remaining launchers 4/4 and remaining missiles 26 (Godot run screenshot)AI & IT tools
AI Tools18 min read

Building Godot game scenes with Claude Code: notes from an air defense game and a canyon flight game

Based on real notes from making an air defense mini game, a penalty shootout, and a canyon flight game with Godot 4.7.2 and Claude Code, this guide covers what to prepare, the first plan, moving Blender models over, automatic checks, and the play-and-fix loop, along with the snags we hit.

The AA-Video-I2V v2.1 leaderboard on Artificial Analysis. 1st Dreamina Seedance 2.5 1152, 2nd Utopai X 1144, 3rd Dreamina Seedance 2.0 1135, 4th MiniMax H3 (768p) 1129, 8th MiniMax H3 Max 1115, 9th Wan 3.0 1111, 11th FLUX 3 1071, 13th Grok Imagine Video 1.5 1050, 14th Gemini Omni Flash 1.1 1043, 17th Kling 3.0 1080p (Pro) 1000AI & IT tools
AI Tools18 min read

How to choose an AI video model: recommendations for live action, animation, film, and motion

Eight AI video models sorted by use: live action, animation and games, film-style camera work, human motion and physics, sound and lip-sync, keeping the same character, and saving money, with leaderboard evidence and results from our own use.

The Output resolution table in the Gemini API docs. The Value column lists 360p, 720p, 1080p, 4k, and the Description column says 360p output resolution, 720p output resolution (default), 1080p output (upscaled), 4K output (upscaled)AI & IT tools
AI Tools20 min read

Eight AI video models compared: specs, prices, and trade-offs as of October 2026

Official names, clip length, resolution, sound, reference inputs, prices, and openness of Gemini Omni 1.1 Flash, Grok Imagine Video 1.5, Wan 3.0, Seedance 2.5, FLUX 3 Video, MiniMax H3, Kling 3.0, and Kling 4.0 in two tables, plus the strong and weak points of each.

The Google DeepMind blog page announcing Genie 4. Date October 27, 2026, title Genie 4: A new generation of world simulation, authors Raia Hadsell, Diego Rivas and Hang Qi, a black Join the Genie 4 waitlist button, and a hummingbird video belowAI & IT tools
AI Tools16 min read

Google DeepMind Genie 4 explained: a real-time world model with sound, and what it means for 3D artists

Using only the official announcement, this guide sums up what is new in Google DeepMind's world model Genie 4, how it differs from Genie 3, who can use it and when, its limits, and what it may mean for Blender and game creators.

Top of the Wan2.1 tutorial in the ComfyUI docs. Title Wan2.1 in ComfyUI: Text-to-Video and Image-to-Video Workflows, a paragraph saying open-sourced by Alibaba in February 2025 under the Apache 2.0 license, and a list of two versions, 14B and 1.3BAI & IT tools
AI Tools21 min read

Turning one character image into video with Wan 2.1: open-source image-to-video vs OpenAI

How the open-source video model Wan 2.1 turns one character image into a short clip, the official ComfyUI steps, and how much graphics card memory it needs. A table compares it with OpenAI, which has shut down its video service.