AI Game Developer: Unity MCP for AI Agents
/Unity-MCP
AI Game Developer is an MCP server and Unity Editor plugin from IvanMurzak/Unity-MCP that lets agents like Claude Code or Cursor create GameObjects and write C# scripts inside a real Unity project. It also runs inside a compiled build, not only the Editor, so an agent can debug at runtime. Reach for it if you already use an MCP-capable agent and want it touching Unity directly; skip it if you don't.
Bringing AI to Unity Development
AI Game Developer is a Unity Editor plugin paired with an MCP server that connects agents such as Claude, Cursor, and Windsurf to a Unity project's scenes, assets, and scripts. It installs as a `.unitypackage` or via OpenUPM, then exposes Unity's Editor APIs as MCP tools an agent can call directly, including a runtime layer that works inside the compiled game itself.
Key Features and Built-in Tools
- ✓More than 70 built-in MCP tools across four categories — Project & Assets, Scene & Hierarchy, Scripting & Editor, and Profiling & Diagnostics — cover everything from creating a GameObject to pulling Editor Profiler memory stats, all available right after install with no extra config.
- ✓The `script-execute` tool compiles and runs C# on the fly through Roslyn, so an agent can test a code change without you switching back to Unity to hit compile.
- ✓Auto-generate Skills reads your OS, Unity version, and installed packages to tailor what the connected agent is told it can do in that specific project.
- ✓Agent choice is provider-agnostic — the README names Anthropic, OpenAI, and Microsoft as interchangeable providers behind whichever client you connect.
- ✓Official extension packages cover Animation, Cinemachine, Input System, Navigation, ParticleSystem, ProBuilder, Splines, Terrain, Tilemap, and Timeline for systems the core tool set doesn't reach.
- ✓Flexible deployment over stdio locally or HTTP remotely, configured per project rather than hard-coded to one transport.
Practical AI Game Development Scenarios
- •Describing a scene change in plain language — arranging primitives, batch-creating objects, tweaking a material — and letting the agent call tools like `gameobject-create` or `assets-material-create` instead of you clicking through the Hierarchy and Inspector.
- •Pointing an agent at Console logs and Profiler stats through the diagnostics tools so it proposes a fix instead of you tracing a stack trace by hand.
- •Writing a game mechanic and its EditMode or PlayMode tests in the same session, using `tests-run` to check the result immediately.
- •Wiring an LLM into a shipped build for NPC dialogue or in-game debugging that reacts to what a player is actually doing, not just what's in the Editor.
Installation Steps for Unity Projects
Two install paths exist, per the README. The GUI route: download the `AI-Game-Dev-Installer.unitypackage` from the latest GitHub release and import it into an open Unity project — double-click the file, or use Assets/Import Package/Custom Package — or run `openupm add com.ivanmurzak.unity.mcp` if you use OpenUPM instead. The CLI route, which the README marks recommended: `npm install -g unity-mcp-cli`, then `unity-mcp-cli install-plugin ./MyUnityProject`; the CLI can also install Unity itself and scaffold a fresh project first if you're starting from nothing. One hard requirement either way: the Unity project path can't contain spaces, so `C:/MyProjects/MyProject` works but `C:/My Projects/MyProject` doesn't.
Connecting and Configuring AI Agents
Pick one agent — Claude Code is the README's recommended choice, alongside GitHub Copilot in VS Code, Cursor, Windsurf, Claude Desktop, and Antigravity — with one real constraint: it needs to be capable of Skills or on-the-fly MCP tool updates. From there, open `Window/AI Game Developer` from Unity's menu bar, then click Auto-generate Skills (the recommended path) or Configure MCP; if your client isn't in the built-in list, the same window shows raw JSON you can paste in by hand. The CLI mirrors this flow: `unity-mcp-cli login` runs a browser OAuth device flow against ai-game.dev and stores the credential locally at `~/.ai-game-dev/credentials.json`, then `unity-mcp-cli open ./MyUnityProject` connects and generates skills automatically. Registering the server manually for a CLI-based agent looks like `claude mcp add ai-game-developer <command>` for Claude Code or `gemini mcp add ai-game-developer <command>` for Gemini CLI, with `<command>` filled in from the OS-specific binary path the README lists for Windows, macOS, and Linux. By default that config aims the agent at a cloud endpoint scoped to this one project; teams can share access instead with an enrollment code passed to `install-plugin --enroll`.
Strengths
- ✓Provider-agnostic by design — Anthropic, OpenAI, and Microsoft-backed agents are named as interchangeable in the README, so switching clients doesn't mean re-wiring the integration.
- ✓Works inside a shipped build, not only inside the Editor, per the README — that's what opens the door to runtime NPC behavior and in-game debugging, not just Editor-time automation.
- ✓Apache-2.0 licensed, and per the repo description, usable with any of the supported agents at no cost.
- ✓Any C# method in the project can become a callable MCP tool with one line, per the repo description — useful if the 70+ built-in tools don't cover something specific to your game.
- ✓An active Discord community for troubleshooting setup and connecting with other developers using the plugin, per the README's own invite link.
Important Requirements and Constraints
- △The Unity project path can't contain spaces — a hard constraint called out in the README, not just a suggestion.
- △AI Game Developer needs an existing Unity Editor and project to connect to; it's a plugin, not something you run standalone.
- △Agent compatibility narrows to clients that support Skills or on-the-fly MCP tool updates — a client without either isn't part of the README's supported list.
- △The CLI's recommended setup runs through a cloud sign-in (`unity-mcp-cli login`, OAuth device flow to ai-game.dev) and aims the agent config at a cloud endpoint scoped to just that project by default, so the easiest onboarding path isn't fully local unless you pass `--no-pin`.
- △No version number, release date, or changelog is included in the facts available here, so there's no documented way to judge how mature the current release is.
Who Benefits from AI Game Developer?
AI Game Developer is for Unity developers who already work inside an MCP-capable agent — Claude Code, Cursor, Windsurf, or similar — and want that agent creating GameObjects, writing C# scripts, and running tests instead of just chatting about code in a side panel. It also fits teams building AI-driven gameplay, like NPCs that call an LLM at runtime, since the plugin's tools work inside a compiled build and not only the Editor. Skip it if your team doesn't use an MCP-aware agent yet, or if you're not comfortable letting an agent modify GameObjects, scripts, and assets directly in your project.
Common Questions About Unity MCP
AI Game Developer works with Claude Code (the README's recommended pick), GitHub Copilot in VS Code, Cursor, Windsurf, Claude Desktop, and Antigravity, plus any other agent that offers Skills support or can pick up new MCP tools on the fly.
AI Game Developer supports this directly — the README notes the plugin runs inside a compiled build, not just the Editor, so an agent can debug problems and react to player behavior while the game is actually running.
AI Game Developer is released under the Apache-2.0 license, as listed on its GitHub repository.
AI Game Developer supports custom tools — per the repo description, any C# method in your project can be turned into a callable MCP tool with a single line, and the README documents adding custom tools in project code.
In Unity, open Edit > Project Settings > AI Game Developer and turn on the toggle that suppresses the update popup for the whole team. It writes to ProjectSettings/AI-Game-Developer-UpdateSettings.asset, so committing that file applies the change to every teammate who pulls it.
AI Game Developer ships a companion CLI, `unity-mcp-cli` (installed via npm), that can install the plugin, sign in via OAuth device flow, open the Unity project, and generate agent skills — a full alternative to using the in-Editor installer.
Alternatives
The problem it solves
Most AI coding agents can read and write files, but they have no way to see or manipulate what's actually running inside the Unity Editor — a GameObject hierarchy, a material's shader properties, the current Play Mode state. Unity-MCP exposes those Editor and runtime internals as MCP tools, so instead of a developer manually clicking through the Hierarchy and Inspector to make a change an agent described, the agent can call `gameobject-create` or `assets-material-create` itself and see the result.
Related repositories
Want a second opinion on Unity-MCP?
Ask an AI that can read this page — one click and you get its take on Unity-MCP.