Unity MCP Server: Connecting AI Assistants to Unity Editor with 330+ Tools

AI-Controlled Unity Editor Without Leaving Your Chat Window

Writing Unity editor scripts, wiring components, sculpting terrain, and managing builds are all discrete tasks that normally require switching between code files, the Unity Editor window, and documentation. Unity MCP Server removes that context switching for AI-assisted workflows: it exposes the Unity Editor surface as a set of MCP tools that an AI assistant like Claude Desktop, Claude Cowork, Cursor, or Windsurf can call directly from the chat window.

The target user is a game developer who already uses a coding assistant and wants the assistant to have live read and write access to the Unity project rather than only reading static code files. The value is clearest for tasks like creating a scene, configuring materials, baking NavMesh, or running a build: tasks that are tedious to do by hand but straightforward to describe in natural language.

The README includes three worked examples that illustrate the scope: a Neon Brick Breaker game built from scratch including neon materials, a physics controller, and UI; a 3D medieval village with terrain sculpting, procedural house building, and environment decoration; and a multi-room castle with lighting adjustments and an FPS walkthrough to verify the result. These examples are described as built by an AI through Unity MCP commands, not hand-written.

Architecture: MCP Server, HTTP Bridge, and Unity Hub CLI

The system has three components. The MCP server in this repository runs as a Node.js process and communicates with your AI assistant over the Model Context Protocol. Inside the Unity Editor, the companion unity-mcp-plugin runs an HTTP API server that the MCP server calls when the AI issues a Unity command. Unity Hub is controlled separately via its command-line interface, which the MCP server uses to list installed editors, manage modules, and set install paths.

When the AI assistant asks to create a GameObject, that request flows from the chat window to the MCP server over MCP, from the MCP server to the Unity Editor plugin over HTTP, and into the running editor instance. The plugin executes the operation and returns the result back along the same path.

On startup the server discovers all running Unity Editor instances automatically. If only one is found, it connects to it immediately. If multiple instances are running, the server exposes a multi-instance selection tool so the AI can target a specific editor.

Setting Up the MCP Server

The server requires Node.js 18 or higher. The package name is anklebreaker-unity-mcp. To start the server after cloning and installing dependencies, the package.json start script runs:

node src/index.js

For development with live reload, the dev script is:

node --watch src/index.js

Before connecting an AI client, the companion unity-mcp-plugin must be installed in the Unity project and the HTTP bridge it runs must be listening. The MCP server will not be able to forward any tool calls to Unity without the plugin. The repository root includes a claude-desktop-config.json file that provides a ready-made configuration snippet for Claude Desktop.

For multi-agent use cases, the MCP server supports a queue monitor and agent action logs exposed through the Multi-Agent category of tools. The stress test script in the tests/ directory (multi-agent-stress-test.mjs) can verify that the queue handles concurrent agent requests without collisions.

Two-Tier Tool System: Core Tools and the Advanced Proxy

With 330+ tools, a flat list would overwhelm most MCP clients, which have limits on how many tools they can display and reason about effectively. Unity MCP Server addresses this with a two-tier design. Roughly 70 core tools are always exposed directly to the client. These cover the most common operations: scene management, basic GameObject manipulation, component access, asset operations, script editing, and play mode control.

The remaining 254 tools are accessed through a single proxy tool called unity_advanced_tool. The client calls unity_advanced_tool with the name of the specific advanced tool it wants to invoke. This keeps the displayed tool count manageable for clients like Claude Desktop and Cowork.

To discover the advanced tools, the README says to call unity_list_advanced_tools, which returns all advanced tools organized by category. This lookup is itself one of the available tools, so the AI assistant can query the full catalog at runtime without needing the categories hardcoded in the system prompt.

Coverage Depth: 30+ Categories Including Shader Graph and MPPM

The tool coverage spans categories beyond the standard Unity Editor surface. The Shader Graph category supports listing, inspecting, creating, and opening Shader Graphs and Sub Graphs, as well as VFX Graphs. The Amplify Shader Editor category provides full graph manipulation including adding, removing, connecting, and disconnecting nodes, only when Amplify is installed.

The MPPM Scenarios category exposes tools for multiplayer playmode scenarios: listing, activating, starting, and stopping scenarios, and retrieving player info. This is specific to Unity's Multi-Player Play Mode package.

The Profiler category starts and stops profiling, collects stats, runs deep profiles, and saves profiler data. The Memory Profiler category adds memory breakdown, top consumer lists, and snapshot capture when the com.unity.memoryprofiler package is installed. The Frame Debugger category lists draw calls and render targets.

For third-party avatar systems, the UMA (Unity Multipurpose Avatar) category creates slots, overlays, and wardrobe recipes from FBX files and manages the Global Library. This level of third-party coverage is unusual for a general-purpose MCP server.

License Terms and Operational Constraints

The package.json license field reads NOASSERTION and says to see the LICENSE file for terms. This means the license is not a standard SPDX identifier. The repository root includes a LICENSE file, but its terms are not summarized in the README. Anyone using Unity MCP Server in a commercial project should read that file directly.

A practical constraint is that the server is only useful when the Unity Editor is running with the HTTP bridge plugin active. It is not a static analysis tool and cannot inspect a Unity project without a live editor connection. This means CI/CD pipelines and headless Unity builds are outside its scope.

Compared to manually writing Unity Editor scripts in C#, the MCP server approach is faster for iterative work but less precise for complex custom behaviors. A handwritten editor script can encapsulate any logic and be version-controlled as part of the project; an AI-driven MCP session produces results on the fly without necessarily leaving a reusable script behind.

Maintenance, Release Cadence, and Version History

The last push to the repository was on 2026-07-27. The repository is not archived. Three releases were published close together: v2.35.6 on 2026-07-27 adding object-reference animation curves, v2.35.5 on 2026-07-24 restoring Unity error messages and adding fail-closed routing, and v2.35.4 on 2026-07-24 adding ProBuilder tools, per-action undo, and lazy discovery. The version history in CHANGELOG.md tracks changes in more detail. The current version is 2.35.6.

Editorial conclusion

Unity MCP Server is a strong fit for developers who want to drive Unity Editor from an AI assistant chat window rather than writing every script and menu interaction by hand. The server covers most of the Unity Editor surface, including categories that third-party plugins like Amplify and MPPM expose. The license file exists but declares an unusual terms arrangement via NOASSERTION in package.json; read LICENSE before using this in a commercial project. Verify that the companion unity-mcp-plugin is installed and the HTTP bridge is running in the editor before starting the MCP server.

Frequently asked questions

Official sources

1. AnkleBreaker-Studio/unity-mcp-server on GitHub
2. Issues
3. Project website
4. README
5. Releases