Unreal MCP Server for Claude Code: What ChiR24/Unreal_mcp Can Control and What to Check Before Setup

ChiR24/Unreal_mcp is an Unreal MCP server that Claude Code can connect to. It controls the Unreal Editor through a native C++ plugin rather than through Python scripting. According to the project README, it covers a wide slice of the editor, from assets and actors to Niagara, Sequencer and graph editing. Node.js is only needed if you use the stdio bridge. Before you install it, check three things: how you will get the plugin into your project, which engine version your build targets, and which other engine plugins it turns on.

This Lab Note reports what the README's features, architecture and prerequisites sections say. It does not judge how well those features work in practice. For plugin installation, follow the README's own install steps, which may change between releases.

What is ChiR24/Unreal_mcp?

The ChiR24/Unreal_mcp README describes an MCP server that "enables AI assistants to control Unreal Engine" through a native C++ Automation Bridge plugin. The project is written in TypeScript and C++ and released under the MIT licence. Its plugin lives in the repository at plugins/McpAutomationBridge. The README's badges show an npm package (unreal-engine-mcp-server), a listing on the MCP Registry, and an engine range of Unreal Engine 5.0 to 5.8.

The README says all operations route through the MCP Automation Bridge plugin. In practice, that means the plugin has to run inside your editor before any tool works.

What does the README say it can control?

The README's features table splits the coverage into ten areas:

Core editor work

• Asset management: browse, import, duplicate, rename and delete assets, and create materials.
• Actor control: spawn, delete, transform, physics, tags and components.
• Editor control: Play-In-Editor (PIE) sessions, camera, viewport, screenshots and bookmarks.
• Level management: load and save levels, streaming and lighting.

Deeper systems

• Animation and physics: Animation Blueprints, state machines, ragdolls, vehicles and constraints.
• Visual effects: Niagara particles, GPU simulations, procedural effects and debug shapes.
• Sequencer: cinematics, timeline control, Movie Render Queue, media, Take Recorder and replay.
• Graph editing: Blueprint, Niagara, Material and Behavior Tree graphs.
• Audio: sound cues, audio components, sound mixes and ambient sounds.
• System: console commands, Unreal Build Tool (UBT), tests, logs, project settings and CVars.

The architecture notes add some useful behaviour. The README says the server starts even when no Unreal editor is connected. It retries the connection handshake with exponential backoff, and it caches asset lookups with a 10-second TTL. It also looks up lights, debug shapes and Sequencer tracks at runtime instead of relying on a fixed list.

All of this comes from the maintainer's own README. The source does not say how reliable each area is, which engine versions each feature was checked against, or how the tools behave on a large project. Treat the table as a list of what the server attempts to do, not proof that every tool works on your project.

How does Claude Code connect to it?

Claude Code is an MCP client, so it can use either of the two transports the README describes:

1. Option A: native MCP transport. The client talks HTTP/SSE straight to the editor plugin. The README says no bridge process is needed, and it covers this route in its Step 4A.
2. Option B: TypeScript stdio bridge. A local Node process speaks stdio to Claude Code and WebSocket to the plugin. The README's Step 1 applies only to this route.

For Option B, the README recommends running the server through npx, with cloning and building the repository as the alternative:

npx unreal-engine-mcp-server

The README's prerequisites say Option B needs Node.js 20.19 or later, and that Node 18 is not supported. The native transport does not need Node. If you would rather not maintain a Node runtime, Option A avoids it. If you pick Option B, run node --version first.

The repository also ships a Dockerfile, and the README's table of contents has a Docker section. This post does not reproduce the client configuration, because ports, paths and token handling can change between releases. Copy them from the README for the version you install.

What safety and access controls does it have?

The README lists these protections:

• Command safety. Dangerous console commands are blocked by pattern-based validation.
• Capability-token auth. Auth is on by default for both the WebSocket and HTTP transports. The server generates a 32-byte secret and stores it at <Project>/Saved/MCP/capability-token. A CapabilityToken set manually in Project Settings overrides the file.
• Metrics rate limiting. The Prometheus metrics endpoint is limited to 60 requests per minute per IP.

The token stops other local processes from talking to the editor without the secret. That is a sensible default for a server that can delete assets and run console commands.

A pattern blocklist is a guard, not a sandbox. It only catches the commands its authors anticipated. An agent that can rename, delete and edit graphs can still damage a project through ordinary tool calls. Before giving any agent editor control:

• Commit the project to version control, using Git LFS or Perforce for binary assets.
• Work on a throwaway branch or a copy of the project.
• Keep Saved/MCP/ out of version control so the token is not committed.

Our guide on setting up Claude Code for game development covers project instructions that keep an agent's edits narrow.

What should you check before installing it?

The features and prerequisites sections answer some setup questions and leave others to the install steps. The table separates the two.

Two practical habits help whatever the answers are. First, look at the .uproject diff after installing, so you can see exactly which plugins were enabled before you commit. Second, plan for engine upgrades: a native editor plugin is part of your project, so check that a working build exists for a new engine version before you upgrade.

How does it compare with other Unreal MCP options?

ChiR24/Unreal_mcp is not the only way to give an agent access to the Unreal Editor. Other open-source projects include groscy/unreal-mcp, remiphilippe/mcp-unreal and the ue-remote command-line tool. Their requirements change often, so this post does not summarise them. Instead, compare each one's README against the same questions:

• Editor side. Does it need a custom compiled plugin, or does it rely on remote interfaces that Epic already ships with the engine?
• Your machine. What runtime does it need (Node, Python or something else), and at which version?
• Engine versions. Which versions does it list, and does an engine upgrade mean rebuilding anything?
• Tool style. Does it offer many typed, area-specific tools, or a few general tools such as running arbitrary scripts?
• Guard rails. Is authentication on by default, and are any commands blocked?

Running arbitrary scripts is flexible, because the agent can do anything the editor's scripting API allows. It also leaves fewer built-in limits on what the agent does. ChiR24's design goes the other way: according to its README, it routes everything through a native plugin with typed tool areas and token auth on by default.

Who is the ChiR24 Unreal MCP server for?

Verdict, based on the README. The C++ Automation Bridge suits Unreal developers who want an agent to reach beyond actors and assets. It fits if you want an agent working in Niagara, Sequencer, Movie Render Queue or Blueprint and Material graphs, and you are comfortable maintaining a native editor plugin in your project. The default token auth and the option to skip Node with the native transport count in its favour.

If you are a solo developer with a Blueprint-only project, or you just want to see whether an agent helps your workflow, read the README's install steps before you start. Confirm you can get a plugin build for your exact engine version without setting up a C++ toolchain. If that turns out to be more work than you want, compare a lighter project against the questions above.

If you work across engines, compare this with the Unity MCP setup for Claude Code or Cursor and the godot-ai MCP server install steps. More engine coverage is under Unreal Engine and coding agents.