Unreal Engine

by remiphilippe

Not rated
GitHub

About

Let AI agents see, build, test, and edit inside Unreal Engine 5.7 — including Blueprints, which are normally opaque binary assets.

Details

Author
remiphilippe
Categories
Other, Developer Tools, AI

Setup

Install Unreal Engine in your MCP client (Claude Desktop, Cursor, Windsurf, and others).

Repository: https://github.com/remiphilippe/mcp-unreal

Follow the installation instructions in the repository README, then restart your MCP client.

MCP (Model Context Protocol) server that gives AI coding agents complete autonomous control over an Unreal Engine 5.7 project. Single Go binary, zero external dependencies.

Build, test, manipulate the editor, edit Blueprints, generate procedural meshes, and look up UE API documentation — all through 49 MCP tools that any MCP-compatible agent can call directly.

# Install go install github.com/remiphilippe/mcp-unreal/cmd/mcp-unreal@latest # Build the documentation index mcp-unreal --build-index # Register with your AI coding tool (see Setup section below)

- Go 1.25+install
- Unreal Engine 5.7— for build/test tools and editor interaction
- Remote Control API plugin— enabled in UE (see
Enabling Remote Control APIbelow)
- MCPUnreal editor plugin— for advanced editor tools (see
plugin/README.md)

git clone https://github.com/remiphilippe/mcp-unreal.git cd mcp-unreal make build

Download fromGitHub Releases— available for macOS, Linux, and Windows on both amd64 and arm64.

The Remote Control API is a built-in UE plugin that provides HTTP access to actor properties and function calls on port 30010. mcp-unreal uses it forget_property,set_property,call_function, andmove_actortools.
- Open your project in Unreal Editor
- Go toEdit > Plugins
- Search for"Remote Control API"
- Check theEnabledcheckbox
- Restart the editor when prompted

You should get a JSON response with the API version. If the connection is refused, the plugin is not enabled or the editor isn't running.

Copy theplugin/directory into your project'sPlugins/folder:

cp -r plugin/ /path/to/YourProject/Plugins/MCPUnreal/

Rebuild the project. The plugin starts an HTTP server on port 8090 when the editor loads. Seeplugin/README.mdfor details.

mcp-unreal works with any MCP-compatible AI coding tool. Below are setup instructions for each.

Note: Replace/path/to/mcp-unrealwith the actual path to your built binary (e.g. the output ofgo installormake build). SetMCP_UNREAL_PROJECTto your.uprojectfile path or project root directory.

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

Config file(.mcp.jsonat project root, or~/.claude.jsonfor global):

{ "mcpServers": { "mcp-unreal": { "type": "stdio", "command": "/path/to/mcp-unreal", "env": { "MCP_UNREAL_PROJECT": "/path/to/MyProject/MyProject.uproject" } } } }

Config file(.cursor/mcp.jsonat project root, or~/.cursor/mcp.jsonfor global):

{ "mcpServers": { "mcp-unreal": { "command": "/path/to/mcp-unreal", "env": { "MCP_UNREAL_PROJECT": "/path/to/MyProject/MyProject.uproject" } } } }

After saving, enable viaCursor Settings > MCP Servers.

codex mcp add mcp-unreal -- /path/to/mcp-unreal

Config file(.codex/config.tomlat project root, or~/.codex/config.tomlfor global):

[mcp_servers.mcp-unreal] command = "/path/to/mcp-unreal" args = [] tool_timeout_sec = 120 [mcp_servers.mcp-unreal.env] MCP_UNREAL_PROJECT = "/path/to/MyProject/MyProject.uproject"

Increasetool_timeout_secforbuild_projectandrun_testswhich can take 60+ seconds.

Requires VS Code 1.99+ with GitHub Copilot Agent Mode enabled.

Config file(.vscode/mcp.jsonat project root):

{ "servers": { "mcp-unreal": { "type": "stdio", "command": "/path/to/mcp-unreal", "env": { "MCP_UNREAL_PROJECT": "${workspaceFolder}" } } } }

Config file(~/.codeium/windsurf/mcp_config.json):

{ "mcpServers": { "mcp-unreal": { "command": "/path/to/mcp-unreal", "env": { "MCP_UNREAL_PROJECT": "/path/to/MyProject/MyProject.uproject" } } } }

Open the Cline panel > MCP Servers icon > Configure tab >Advanced MCP Settings, then add:

{ "mcpServers": { "mcp-unreal": { "command": "/path/to/mcp-unreal", "env": { "MCP_UNREAL_PROJECT": "/path/to/MyProject/MyProject.uproject" } } } }

Config file(~/.config/zed/settings.json):

{ "context_servers": { "mcp-unreal": { "command": "/path/to/mcp-unreal", "env": { "MCP_UNREAL_PROJECT": "/path/to/MyProject/MyProject.uproject" } } } }

For best results, add the following to your project's system prompt (e.g.CLAUDE.md,.cursorrules,.codex/instructions.md, or equivalent). This tells the AI agent how to use the MCP tools effectively.

## Unreal Engine MCP Tools This project uses mcp-unreal for AI-driven Unreal Engine development. The following MCP tools are available: ### Workflow Guidelines 1. Always check status first: Call status to verify the editor and plugin connections before attempting editor operations. 2. Look up docs before writing UE code: Use lookup_class to get class references (inheritance, properties, functions) and lookup_docs for API patterns before writing C++ or Blueprint logic. 3. Build-test cycle: After editing C++ files, call build_project to compile, then run_tests to verify. Read build errors carefully — use lookup_docs to understand UE APIs. 4. Actor workflow: Use get_level_actors to discover existing actors, spawn_actor to create new ones, move_actor to position them, and set_property/get_property to read/write properties. 5. Blueprint workflow: Use blueprint_query with operation list to find Blueprints, then list_variables/list_functions/list_nodes to inspect them. Use blueprint_modify to make changes, and always compile after modifications. 6. Material workflow: Use material_ops with create to make new materials, get_parameters to inspect existing ones, and set_parameter or set_texture to modify them. 7. Save your work: The editor does not auto-save. After making changes, use level_ops with save_level or remind the user to save. 8. PIE testing: Use pie_control with operation start to begin a Play In Editor session, status to verify it started, and stop to end it. Use capture_viewport with include_ui=true during PIE to see the game view with HUD/menus. Use player_control to move the player pawn (teleport, set_rotation) or reposition the editor viewport camera (get_camera, set_camera). ### Tool Modes - Headless tools (build_project, run_tests, cook_project, etc.) do NOT require the editor to be running. They invoke UnrealEditor-Cmd directly. - Editor tools (actors, blueprints, materials, etc.) require the Unreal Editor to be open with the MCPUnreal plugin loaded. - Documentation tools (lookup_docs, lookup_class) are always available — use them liberally. ### Object Paths Many tools require UE object paths. The format is: - Actor paths: /Game/Maps/MapName.MapName:PersistentLevel.ActorName - Asset paths: /Game/Folder/AssetName - Engine assets: /Engine/BasicShapes/Cube Use get_level_actors to discover actor paths, and search_assets to find asset paths.

Tip: For Claude Code, save this asCLAUDE.mdin your project root. For Cursor, save as.cursorrules. For Codex, save as.codex/instructions.md.

- macOS:/Users/Shared/Epic Games/UE_5.7/Engine/Binaries/Mac/UnrealEditor-Cmd
- Windows:C:\Program Files\Epic Games\UE_5.7\Engine\Binaries\Win64\UnrealEditor-Cmd.exe
- Linux:/opt/UnrealEngine/Engine/Binaries/Linux/UnrealEditor-Cmd

┌───────────────────────┐ │ UE 5.7 Editor │ │ │ │ ┌─────────────────┐ │ │ │ Remote Control │ │ ┌────►│ │ API (port 30010)│ │ │ │ │ (built-in) │ │ │ │ └─────────────────┘ │ ┌──────────────┐ stdio ┌──────────────┐│ │ │ │ AI Coding │ JSON-RPC │ ││ │ ┌─────────────────┐ │ │ Agent │◄────────────►│ mcp-unreal │├────►│ │ MCPUnreal │ │ │ (Claude Code │ │ (Go binary) ││ │ │ Plugin (port │ │ │ Cursor, etc)│ │ ││ │ │ 8090) │ │ └──────────────┘ │ 49 tools │┘ │ │ • Actors │ │ │ doc index │ │ │ • Blueprints │ │ │ │ │ │ • Materials │ │ │ ┌──────────┐ │ │ │ • PCG / GAS │ │ │ │ Headless │ │ │ │ • Niagara │ │ │ │ exec.Cmd │─┼──────│──│ • Mesh ops │ │ │ └──────────┘ │ │ │ • Levels │ │ │ │ │ └─────────────────┘ │ │ ┌──────────┐ │ │ │ │ │ Bleve │ │ └───────────────────────┘ │ │ Doc Index│ │ │ └──────────┘ │ ┌───────────────────────┐ │ │──────│ docs/ │ └──────────────┘ │ ├── ue5.7/ │ │ ├── realtimemesh/ │ │ └── index.bleve │ └───────────────────────┘

- Headlessexec.CommandinvokesUnrealEditor-Cmdfor builds, tests, and cooking. No editor needed.
- Remote Control API— HTTP PUT tolocalhost:30010for property access and function calls. Built into UE.
- MCPUnreal Plugin— HTTP POST tolocalhost:8090for Blueprint editing, asset queries, mesh ops, and other deep editor internals.

SeeIMPLEMENTATION.mdfor the full architecture document.

Build the local documentation search index:

This indexes markdown files fromdocs/ue5.7/anddocs/realtimemesh/, plus your project'sCLAUDE.md. The index is stored at./docs/index.bleve(configurable viaMCP_UNREAL_DOCS_INDEX).

To add custom documentation, place markdown files in thedocs/directory and rebuild the index. Seedocs/README.md.

Once registered, your AI agent can autonomously:

You: "Build the project, fix any errors, and run the tests" Agent calls: 1. status() → editor + plugin online 2. build_project() → finds 2 compile errors 3. lookup_class("AActor") → looks up the correct API 4. (edits the C++ source) 5. build_project() → clean build 6. run_tests(filter: "Game") → 14/14 tests pass
You: "Add a Niagara fire effect to the torch actor" Agent calls: 1. get_level_actors(name_filter: "Torch") → finds actor path 2. lookup_class("UNiagaraComponent") → checks API 3. niagara_ops(operation: "spawn_system", system_path: "/Game/FX/NS_Fire", ...) 4. niagara_ops(operation: "set_parameter", parameter_name: "SpawnRate", value: 50)
You: "Create a PCG scatter for placing rocks on the landscape" Agent calls: 1. lookup_docs("PCG scatter landscape") → reads PCG docs 2. pcg_ops(operation: "list_components") → finds existing PCG actors 3. spawn_actor(class: "Actor", name: "RockScatter") 4. pcg_ops(operation: "add_node", node_type: "SurfaceSampler") 5. pcg_ops(operation: "execute") → generates rocks
# Build make build # Run all tests with race detector make test # Go lint make lint # C++ format check (requires clang-format) make cpp-fmt-check # Run all checks (Go format, C++ format, vet, lint, tests) make check

The MCPUnreal plugin includes automation tests that run inside the UE editor headlessly. These require UE 5.7 installed locally and arenotpart of CI — they are opt-in for local development.

# Build the test project (first run ~2-5 min, incremental after) make test-cpp-build # Build and run all MCPUnreal. automation tests make test-cpp

This uses a minimal UE project intest-project/that symlinks the plugin source and runs allMCPUnreal.tests viaUnrealEditor-Cmd -nullrhi.

SeeCONTRIBUTING.mdfor development setup, coding standards, and how to add new tools.

Unreal Engine is a trademark of Epic Games, Inc. This project is not affiliated with or endorsed by Epic Games.

Premium MCP server for Godot game engine with 84 AI-powered tools for scene editing, scripting, animation, tilemap, shader, input simulation, and runtime debugging.

An AI-powered MCP server for Roblox Studio development, featuring advanced NLP, semantic analysis, and multi-turn conversation capabilities.

Powerful tool for the Unity Editor that gives AI Agents ability to perform any action using Unity Editor API, like modification of scripts, scenes, prefabs, assets, configuration and more.

Perform actions in the Unity Editor for game development using AI clients.

An MCP server that allows AI assistants to programmatically interact with Unity development projects.

A unified server to control Blender and Unreal Engine via AI agents.

An unofficial Unreal Engine plugin that acts as an MCP server, allowing AI tools to remotely control the engine.

AI-powered no-code app builder with 17 MCP tools — create projects, generate pages from natural language, AI text/image generation (GPT, Claude, Gemini, 14+ models), page CRUD, workflow execution, publish & version control. SSE transport, API key auth.

Unified MCP & skill management gateway with progressive disclosure. Manages multiple MCP servers as Agent Apps, loading tool schemas on demand for 99% context token savings. Shared across Claude Code, Codex, OpenCode and more.

Cyber Host Artificial Intelligence (C.H.A.I) is Autonomous penetration testing MCP (Model Context Protocol) server with an integrated AI decision engine, multi-provider LLM support, and an extensible plugin architecture.

No reviews yet — be the first

Sign in to leave a review

Use Google, GitHub, or an email account so ratings stay tied to real people.

Email sign in

No reviews posted yet.