FunplayAI/funplay-unity-mcp

Window → Package Manager → + → Add package from git URL → https://github.com/FunplayAI/funplay-unity-mcp.git
openupm add com.gamebooom.unity.mcp  # or the OpenUPM scoped registry
Start the server via Funplay → MCP Server, then use the built-in one-click client configuration

Usage

• •Give a one-sentence prompt like 'Build a snake game with a 10x10 grid, food spawning, score UI, and game-over screen'
• •Verify with safe calls first: get_scene_info, read unity://project/context, execute_code
• •Edit exposed tool profiles under Funplay → Tool Exposure, and execute_code safety defaults under Funplay → MCP Settings

Key features

• ✓156 tools across 35 modules with structured {success, message, data} returns and stable instanceId chaining
• ✓execute_code: Roslyn-first in-memory C# execution on the editor thread with auto-Undo and tracked changelogs
• ✓Play-mode automation loop (enter play, simulate mouse/keyboard, capture screenshots, inspect logs, validate)
• ✓Profiler, memory snapshot, frame debugger tools plus project resources, prompts and interaction history
• ✓Vendor agnostic — works with Claude Code, Cursor, LM Studio, Windsurf, Codex, VS Code Copilot, etc.

Best for

Unity developers who want AI agents to build scenes, scripts, UI and validate behaviour directly inside the running editor.

Caveats

• ⚠Editor-only package — adds no runtime components to built games
• ⚠execute_code safety guards are enabled by default but are not a complete sandbox; no approval toggle exists for exposed MCP tools
• ⚠Requires Unity 2022.3+ and Newtonsoft.Json