You've heard about MCP. You've seen the demos. Maybe you've read our complete guide to AI-assisted level design and felt a bit overwhelmed by the depth. That guide covers everything. This one covers just enough to get you running.

This is the fastest path from "I've never touched MCP" to "I just spawned an actor in Unreal using natural language." Fifteen minutes. Let's go.

What Is MCP (the 30-Second Version)

MCP stands for Model Context Protocol. It's an open standard that lets AI assistants talk to external tools. Instead of the AI generating code that you copy and paste into your editor, MCP lets the AI directly interact with your editor in real time.

For Unreal Engine, that means you can type something like "place a directional light and angle it for late afternoon" into your AI client, and the light appears in your viewport. No scripting. No manual property editing. The AI calls the right editor functions through the MCP connection.

Think of MCP as a translator. Your AI speaks natural language. Unreal Engine speaks C++ API calls. MCP sits between them and handles the translation.

Why Should You Care?

Three reasons:

• Speed — repetitive editor tasks that take minutes of clicking become single sentences
• Accessibility — you don't need to know every submenu and property panel in the editor
• Iteration — describe changes in plain English, see results instantly, undo if you don't like them

It's not magic, and it won't design your game for you. But it will handle the mechanical parts of editor work so you can focus on creative decisions.

Prerequisites

Before we start, make sure you have:

• Unreal Engine 5.3 or later installed and working
• A UE5 project open (any project works, even a blank template)
• An MCP-compatible AI client — we'll cover Claude Code, Cursor, and Windsurf below
• The Unreal MCP Server plugin downloaded from your StraySpark account

That's it. No Python environments to configure. No Docker containers. No cloud API keys. Everything runs locally on your machine.

A Note on System Requirements

The MCP server itself is lightweight — it adds negligible overhead to your Unreal project. The AI client runs separately. If your machine can run UE5, it can run MCP alongside it without issues.

Step 1: Install the Plugin

The Unreal MCP Server installs like any other UE5 plugin. Here's the process:

1. Close Unreal Engine if it's running
2. Copy the plugin folder into your project's Plugins directory. If you don't have a Plugins folder at the root of your project, create one
3. Open your project in Unreal Engine
4. Go to Edit > Plugins and search for "MCP Server"
5. Confirm it's enabled (it should be by default after copying)
6. Restart the editor when prompted

After the restart, the MCP server starts automatically in the background. You'll see a brief log entry in the Output Log confirming it's running. That's the Unreal side done.

Verifying the Server Is Running

If you want to double-check, open the Output Log (Window > Developer Tools > Output Log) and filter for "MCP." You should see a message indicating the server is listening for connections. If you don't see it, make sure the plugin is enabled in the Plugins menu and restart the editor.

Common Installation Issues

A few things that can trip you up at this stage:

• Wrong engine version — the plugin needs UE5.3 or later. If you're on an older version, you'll get a compatibility warning at startup
• Plugin folder structure — make sure the plugin's .uplugin file is inside YourProject/Plugins/UnrealMCPServer/ (not nested deeper)
• Firewall prompts — the server communicates over a local network port. If your firewall asks, allow the connection. It's all local traffic, nothing leaves your machine

Step 2: Configure Your AI Client

The MCP server is running. Now you need to point your AI client at it. This step varies slightly depending on which client you use, but the core idea is the same: tell the client where the MCP server lives.

Option A: Claude Code

Claude Code is Anthropic's command-line interface for Claude. If you're comfortable with a terminal, this is the most straightforward option.

1. Open your terminal
2. Navigate to your Unreal project directory
3. Add the MCP server to your Claude Code configuration:

claude mcp add unreal-engine -- /path/to/unreal-mcp-server

Replace /path/to/unreal-mcp-server with the actual path to the MCP server executable that shipped with the plugin. You'll find it in the plugin's Binaries folder.

4. Start Claude Code:

claude

5. Claude Code will automatically detect the MCP server and connect to it. You'll see the available tools listed when the session starts.

That's it. Claude Code is ready to talk to Unreal.

Option B: Cursor

Cursor is an AI-powered code editor that supports MCP natively.

1. Open Cursor
2. Go to Settings > MCP Servers
3. Click Add Server
4. Enter the server details:
   • Name: Unreal Engine
   • Command: path to the MCP server executable
5. Save and restart Cursor

Once connected, you can use Cursor's AI chat to send commands to Unreal Engine directly. The MCP tools appear in Cursor's tool palette.

Option C: Windsurf

Windsurf (Codeium's AI IDE) also supports MCP.

1. Open Windsurf
2. Navigate to the MCP configuration in settings
3. Add a new server pointing to the Unreal MCP Server executable
4. Save and reconnect

The configuration pattern is identical across clients: point the client at the server, and the MCP protocol handles the rest. If your preferred tool supports MCP, it will work with the Unreal MCP Server.

How Do I Know It's Connected?

Every client shows connected MCP servers differently, but the key indicator is the same: you should see Unreal-specific tools available in the AI's tool list. If you ask the AI "what tools do you have access to?" and it mentions things like actor spawning, material editing, or level querying, you're connected.

Step 3: Your First Command

Everything is installed and connected. Time to actually do something.

Open your Unreal project so you have an empty (or any) level visible in the viewport. Then, in your AI client, type:

"Spawn a cube at the origin"

The AI will call the appropriate MCP tool to create a static mesh actor with a cube mesh at world position (0, 0, 0). You'll see it appear in your viewport.

That's your first MCP command. From here, you can start experimenting.

Try These Next

Once the cube is there, try a few more commands to get a feel for what's possible:

"Move the cube up 200 units"

"Change the cube's material to a red metallic surface"

"Add a point light 300 units above the cube with warm white color"

"Duplicate the cube and place the copy 500 units to the right"

"Select all actors in the scene and list their names"

Each of these maps to specific MCP tool calls. The AI figures out which tools to use based on your description. You don't need to know the tool names or parameters — that's the whole point.

Understanding What Happened

When you said "spawn a cube at the origin," here's what actually happened behind the scenes:

1. Your AI client sent your message to the AI model (Claude, in this case)
2. The AI interpreted your request and decided to call the spawn_actor MCP tool
3. The MCP client sent the tool call to the Unreal MCP Server
4. The server executed the corresponding editor command inside Unreal
5. The result was sent back to the AI, which confirmed the action to you

All of this happened in a second or two. And critically, the action is recorded in Unreal's undo history. If you don't like what the AI did, Ctrl+Z works normally.

Step 4: Exploring Tool Presets

The Unreal MCP Server includes 207 tools across 34 categories. That's powerful, but it can also be overwhelming — both for you and for the AI. More available tools means the AI has more choices to sort through, which can occasionally lead to it picking a suboptimal tool.

That's where tool presets come in. Presets are curated subsets of tools tailored to specific workflows:

• Level Design — actors, transforms, landscape, foliage. Perfect for blockouts and environment work
• Visual Polish — materials, lighting, post-processing, atmosphere. Use this when you're refining the look of a scene
• Blueprint Development — Blueprint creation, component management, property editing. For gameplay logic work
• Cinematics — Sequencer, cameras, animation. When you're building cutscenes or trailers
• Full Access — all 200+ tools. Use this when you need everything, or when you're doing mixed-discipline work

How to Switch Presets

You can tell the AI which preset to use:

"Switch to the Level Design preset"

Or you can configure the default preset in the plugin settings within Unreal. For your first session, stick with Level Design — it has the most immediately useful tools and keeps the AI focused.

Why Presets Matter

We've found that AI accuracy improves noticeably when you use the right preset. If you're doing lighting work with the Full Access preset, the AI has 200+ tools to choose from. With the Visual Polish preset, it has a focused set of lighting-specific tools and makes better decisions. Think of it like giving someone the right toolbox for the job instead of handing them an entire hardware store.

Common First-Timer Mistakes

We've onboarded a lot of developers onto MCP at this point. Here are the mistakes we see most often and how to avoid them.

Mistake 1: Being Too Vague

"Make the level look good" gives the AI almost nothing to work with. It might do something, but the result probably won't match what you had in mind.

Instead, be specific: "Add volumetric fog with density 0.02 and a warm amber tint" or "place 6 point lights evenly spaced around the room at a height of 300 units."

The AI is excellent at executing precise instructions. It's less reliable at interpreting creative direction. Give it numbers, positions, and concrete descriptions.

Mistake 2: Trying to Build Everything in One Prompt

"Build me a medieval village with 20 houses, a market square, surrounding walls, a gate, NPCs walking around, and dynamic day-night lighting" is a lot to ask in a single request.

Break it down. Work in passes, just like you would with manual level design:

1. First pass: block out the major structures
2. Second pass: add environmental details
3. Third pass: lighting and atmosphere
4. Fourth pass: polish and iteration

Each pass can be a few focused prompts. This gives you checkpoints where you can review and adjust before moving on.

Mistake 3: Forgetting About Undo

The AI will occasionally place something wrong or misinterpret your request. That's fine. Ctrl+Z is your best friend. Every MCP action goes through Unreal's standard transaction system, so undo works exactly as you'd expect.

Don't be afraid to experiment aggressively. If you don't like the result, undo it and try rephrasing your request.

Mistake 4: Not Using Context

The MCP server provides 12 context resources that let the AI understand your current scene. If the AI seems confused about what's already in your level, try asking it to review the scene first:

"List all actors in the current level"

"What materials are currently in use?"

"Describe the current lighting setup"

These queries populate the AI's understanding of your scene, leading to better subsequent commands.

Mistake 5: Fighting the Tool Instead of Using It as a Starting Point

MCP is best used as an accelerator, not a replacement for manual work. Let the AI handle the initial placement and bulk operations, then switch to manual editor work for fine-tuning. The best workflows combine AI speed with human creative judgment.

If you find yourself writing increasingly long prompts trying to get the AI to place something in exactly the right spot, it's probably faster to let the AI get close and then drag the actor into position manually.

Understanding the 200+ Tools

You don't need to memorize all 200+ tools. In fact, you don't need to know any of them by name. The AI handles tool selection for you based on your natural language description. But it helps to know what categories exist so you understand the boundaries of what's possible.

Here's a high-level overview of the 32 tool categories:

• Actor Management — spawning, deleting, duplicating, renaming actors
• Transform — position, rotation, scale operations
• Materials — creating, assigning, and modifying materials
• Lighting — all light types, shadows, global illumination settings
• Landscape — terrain creation and modification
• Foliage — foliage painting and management
• Blueprints — class creation, component management, property editing
• Sequencer — timeline, tracks, keyframes
• Camera — camera actors, view targets, filmback settings
• Post-Processing — volumes, bloom, exposure, color grading
• Audio — sound actors, attenuation, spatialization
• World Settings — game mode, gravity, world composition

And more. The full list is in the documentation, but again, you don't need to study it. Just describe what you want in plain language and the AI will find the right tools.

From First Command to Real Workflow

Once you're comfortable with basic commands, here are some practical workflows to try:

Rapid Level Blockout

Use the Level Design preset and describe your level layout in plain language. Start with the overall space, then add structural elements, then details. A basic blockout that would take 30 minutes of manual BSP work can be done in 5 minutes of conversation.

Lighting Iteration

Switch to the Visual Polish preset and experiment with lighting setups. Ask the AI to try different configurations: "try three-point lighting on the main character," "switch to dramatic side lighting," "add rim light from behind." Each change takes seconds instead of minutes of property panel clicking.

Scene Auditing

Ask the AI to analyze your existing scene: "find any overlapping actors," "list all lights with default intensity," "show me actors that aren't in any folder." This kind of audit is tedious by hand but instant through MCP.

Batch Property Editing

Need to change a property on dozens of actors at once? "Set all point lights in the scene to cast shadows" or "change the mobility of every static mesh to Movable." These batch operations are where MCP saves the most time compared to manual work.

What Comes Next

You've installed the plugin, connected your AI client, and run your first commands. Here's where to go from here:

• Read the complete MCP guide for deeper coverage of workflows, context resources, and advanced techniques
• Explore tool presets by trying each one with a relevant task
• Build a small scene from scratch using only AI commands to build muscle memory
• Try the Blender MCP Server if you also work in Blender — the same concepts apply with 212 tools across 22 categories
• Check out the Complete Toolkit bundle if you want the full StraySpark toolset

The learning curve for MCP is genuinely short. Most developers go from first install to productive workflow within a single session. The hard part isn't learning the tools — it's unlearning the habit of doing everything manually.

Welcome to AI-assisted development. Your editor just got a lot more conversational.