Unity AI Bridge

What butterlatte-zhang/unity-ai-bridge MCP server does

The butterlatte-zhang/unity-ai-bridge MCP server connects an AI coding environment to the Unity Editor. Its tool set covers editor workflows that are not available through ordinary file editing, including scene management, GameObject and component changes, asset operations, prefabs, scripts, selection state, reflection calls, screenshots, runtime queries, console logs, profiling, package management, light probes, and test execution.

The repository describes 65 tools in 15 categories. This includes seven scene tools, eleven each for GameObjects and assets, five prefab tools, four script tools, and smaller groups for editor state, runtime access, profiling, packages, light probes, and tests. The tool names and categories indicate that the bridge can support both content editing and iterative validation inside a Unity project.

How it works

The integration has two client-facing paths: Skill mode for direct CLI use and MCP mode for protocol-based clients. Both use the same Unity plugin and the same file-based request/response channel. A Python component writes requests to disk, the Unity package polls for them, runs the corresponding registered method, and writes a response back.

This design avoids open ports, firewall configuration, and socket-specific connection handling. The README states that file polling is intended to remain reliable during Unity domain reloads, Play Mode transitions, recompilation, and Editor restarts. The stated polling interval is approximately 100 milliseconds.

A typical testing loop can combine script updates with compilation-log inspection, Play Mode control, runtime method calls, runtime state queries, and screenshots. The repository describes this as an act, wait, observe, judge, and repeat workflow for automated playtesting.

Setup and configuration

The Unity side is installed from the repository through Unity Package Manager using this Git URL:

https://github.com/butterlatte-zhang/unity-ai-bridge.git?path=Packages/com.aibridge.unity

The alternative is to copy Packages/com.aibridge.unity into the Unity project's Packages/ directory. IDE integration requires copying the repository's .claude/ directory to the project root and then following the setup guide for the selected client. The repository points to docs/SETUP.md for the client-specific configuration steps.

Supported integrations listed in the material are Claude Code in Skill mode, Cursor, GitHub Copilot, Windsurf, and Claude Desktop in MCP mode. The project targets Unity 2022.3 LTS or newer.

Tools and capabilities

The butterlatte-zhang/unity-ai-bridge MCP server includes tools for:

• Opening, creating, saving, listing, activating, and unloading scenes.
• Finding, creating, modifying, duplicating, parenting, and destroying GameObjects and components.
• Finding, reading, changing, moving, copying, deleting, and creating asset content.
• Creating, opening, saving, closing, and instantiating prefabs.
• Reading, creating, updating, deleting, and executing scripts.
• Reading console logs, taking screenshots, and controlling editor selection or application state.
• Calling reflected methods and querying or invoking runtime behavior.
• Capturing profiler snapshots, streams, frame hierarchies, hot paths, and garbage-collection allocations.
• Listing, searching, adding, and removing Unity packages.
• Generating, analyzing, baking, clearing, and configuring light probes.
• Running Unity tests.

Custom tools can be added with the repository's BridgeTool attribute; the README says these methods are discovered, serialized, and documented automatically.

Limitations and notes

The bridge requires a Unity project and the Unity Editor package; it is not a standalone way to control arbitrary applications. File polling introduces roughly 100 ms of IPC latency. The repository describes the implementation as a self-contained Python standard-library CLI/MCP server plus a C# Unity package, with no external Python, Node.js, or npm dependencies. No environment variables or API credentials are specified in the provided material.