Unreal AI Connection

What NAJEMWEHBE/unreal-ai-connection MCP server does

NAJEMWEHBE/unreal-ai-connection MCP server gives an MCP-compatible client access to a running Unreal Engine editor. It is intended for editor automation rather than gameplay runtime control. The available surface covers project information, asset registry queries, asset changes, Blueprint inspection and compilation, widget authoring, specialized asset inspection, and arbitrary Unreal Python execution.

The repository describes a native C++ Unreal plugin paired with a thin Python bridge. The plugin handles editor operations inside Unreal, while the bridge presents those operations through MCP. The project reports support for Unreal Engine 5.7 as its last-known-good version and includes a prebuilt UE 5.6 release. Other engine versions are best-effort and untested.

How it works

An MCP client sends a tool call to the Python bridge over stdio. The bridge forwards the request as framed JSON-RPC through a TCP connection to the Unreal plugin at 127.0.0.1:18888. The plugin then calls Unreal Editor APIs and returns the result through the same path. Because the editor endpoint listens only on the local loopback address, the documented setup does not expose the editor to other machines over the network.

The bridge includes several Python-oriented workflows. Agents can execute multiline code in the editor's unreal.* interpreter, run a Python file, apply code to the current actor or asset selection, or use a persistent Python session whose variables survive between calls. A reset operation clears that persistent session state.

Setup and configuration

The Unreal plugin must be installed separately in the project’s Plugins/ directory, and the editor must be launched so it can bind its local socket. The MCP client then needs to be configured to start the repository’s Python bridge script. The README provides client-specific configuration paths for Claude Desktop, Cursor, Windsurf, Cline, and other MCP clients, plus recipes for Windows users who need the py launcher instead of python3.

The documented examples use an absolute path to bridge/unreal_ai_connection_bridge.py; that path must be replaced with the location on the host running the bridge. The project also requires a compatible Python installation; its repository badge specifies Python 3.11 or newer. No API key or remote service is described.

Tools and capabilities

NAJEMWEHBE/unreal-ai-connection MCP server includes tools for:

• Summarizing the project, including its engine version, enabled plugins, and asset count.
• Finding, inspecting, moving, renaming, duplicating, and deleting assets, with redirector handling and reference checks.
• Creating data tables, data assets, and Blueprint assets.
• Reading and compiling Blueprints, including variables, graphs, and parent classes.
• Inspecting Widget Blueprints and Widget Trees, then changing a tree’s root, children, or properties.
• Inspecting animation Blueprints, montages, static and skeletal meshes, physics assets, Niagara systems, landscapes, data tables, and curves.
• Running arbitrary or persistent Unreal Python code in the editor.

The broader README also describes authoring and automation coverage beyond the individual tools listed in the supplied material, including actors, levels, material graphs, Niagara effects, cinematics, and render-queue workflows.

Limitations and notes

The repository is explicitly deprecated. Active development stopped in July 2026, and the maintainer states that no further features, fixes, releases, or issue or pull-request responses are promised. The code remains available under the MIT license, and forks are permitted.

UE 5.7 is the officially supported and tested target identified by the project. UE 5.6 is covered by a named prebuilt release, while other versions are not promised. Source builds for UE 5.3 and older may require manually selecting an engine-appropriate MSVC compiler version. The tool requires a running editor with the plugin installed; configuring only the stdio bridge is not sufficient.