wordle/.claude/skills/gds-game-architecture/engine-mcps.yaml

# Engine MCP Servers - Game engine MCP integrations for AI-assisted development
# Used during Step 3 (engine selection) to recommend relevant MCP servers
#
# ⚠️ CRITICAL: All MCP availability and versions MUST be verified via WebSearch
# This file provides: known MCPs, search terms for discovery, and capability descriptions
# MCP ecosystem evolves rapidly - always verify repos are active and maintained
#
# MCP (Model Context Protocol) servers give AI agents direct access to game engine
# features like scene inspection, asset queries, and live debugging - dramatically
# improving AI-assisted game development.

engine_mcps:
  unity:
    engine: "Unity"
    mcps:
      mcp_unity:
        name: "MCP Unity (CoderGamester)"
        repo: "CoderGamester/mcp-unity"
        search_term: "CoderGamester mcp-unity github"
        description: "Full-featured MCP bridge for Unity Editor with 30+ tools for scene, asset, and component management"
        capabilities:
          - "Create, load, save, and delete scenes"
          - "Create, select, update, duplicate, delete, and reparent GameObjects"
          - "Add and update components with property access"
          - "Create, assign, and modify materials"
          - "Add assets to scenes, manage packages"
          - "Run Unity Test Runner tests"
          - "Recompile scripts, execute menu items"
          - "Batch execute multiple commands atomically"
          - "Access Unity console logs and output"
        install_type: "Unity Package Manager (git URL) + Node.js MCP server"
        install_steps:
          - "Install Node.js 18+"
          - "In Unity: Window > Package Manager > + > Add package from git URL"
          - "URL: https://github.com/CoderGamester/mcp-unity.git"
          - "In Unity: Tools > MCP Unity > Server Window > Configure"
          - "Click Start Server"
        requirements: "Unity 6+, Node.js 18+, npm 9+"
        supported_clients: ["Claude Desktop", "Claude Code", "Cursor", "Windsurf", "Codex CLI", "GitHub Copilot"]
        license: "MIT"
        benefit_summary: "AI agents get 30+ tools to directly inspect and modify your Unity scenes, GameObjects, components, and materials - no copy-pasting descriptions needed."

      unity_mcp_coplay:
        name: "Unity MCP (CoplayDev)"
        repo: "CoplayDev/unity-mcp"
        search_term: "CoplayDev unity-mcp github"
        description: "AI bridge for Unity Editor with natural language control, asset and scene management, and batch operations"
        capabilities:
          - "Natural language control of Unity Editor"
          - "Manage assets, scenes, materials, and scripts"
          - "Batch execute operations (10-100x faster than individual calls)"
          - "Two-way communication between AI and Unity"
        install_type: "Unity Package Manager (git URL) or Unity Asset Store or OpenUPM"
        install_steps:
          - "In Unity: Window > Package Manager > + > Add package from git URL"
          - "URL: https://github.com/CoplayDev/unity-mcp.git?path=/MCPForUnity"
          - "Configure your MCP client to connect"
        requirements: "Unity 2021.3 LTS+"
        supported_clients: ["Claude Desktop", "Cursor", "Windsurf"]
        license: "MIT"
        benefit_summary: "Broad Unity version support (2021.3+) with natural language control and fast batch operations for AI-driven scene building."

      unity_mcp_official:
        name: "Unity MCP (Official)"
        repo: "Unity AI Assistant package (com.unity.ai.assistant)"
        search_term: "Unity MCP official Unity AI Assistant docs"
        description: "Unity's first-party MCP server, bundled with the Unity AI Assistant, exposing the Editor to external AI clients"
        capabilities:
          - "Manage scenes and GameObjects"
          - "Run asset operations"
          - "Edit scripts"
          - "Read Unity console output"
          - "Automate in-editor tasks"
        install_type: "Bundled in the Unity AI Assistant package - no separate server"
        install_steps:
          - "Open Edit > Project Settings > AI > Unity MCP"
          - "The MCP bridge starts automatically with the Editor"
          - "Approve direct connections from your AI client"
        requirements: "Unity 6+ with the AI Assistant package; paid Unity AI subscription (~$10/month - verify current pricing)"
        supported_clients: ["Claude Code", "Cursor"]
        license: "Proprietary (Unity)"
        docs: "https://docs.unity3d.com/Packages/com.unity.ai.assistant@2.8/manual/integration/unity-mcp-overview.html"
        note: "Requires a paid Unity AI subscription. The free open-source MCPs (mcp_unity, unity_mcp_coplay) cover the same workflow - prefer those unless the user already pays for Unity AI."
        benefit_summary: "First-party MCP built into the Unity Editor, but gated behind a paid Unity AI subscription; the open-source alternatives are free."

  unreal:
    engine: "Unreal Engine"
    mcps:
      unreal_mcp:
        name: "Unreal MCP (chongdashu)"
        repo: "chongdashu/unreal-mcp"
        search_term: "chongdashu unreal-mcp github"
        description: "Enables AI assistants to control Unreal Engine through natural language via C++ plugin and Python server"
        capabilities:
          - "Create, delete, and transform actors in levels"
          - "Query actor properties and list level contents"
          - "Create Blueprint classes with custom components"
          - "Configure Blueprint properties and compile"
          - "Add event nodes, function calls, and variables to Blueprint graphs"
          - "Focus viewport on actors, control camera orientation"
          - "Create input action mappings"
        install_type: "Unreal plugin (C++) + Python MCP server"
        install_steps:
          - "Copy UnrealMCP plugin to your project's Plugins folder"
          - "Enable via Edit > Plugins > search 'UnrealMCP'"
          - "Generate Visual Studio files and build (Development Editor)"
          - "Install Python 3.12+ and set up the Python server per Python/README.md"
          - "Configure MCP client with provided JSON config"
        requirements: "Unreal Engine 5.5+, Python 3.12+"
        supported_clients: ["Claude Desktop", "Cursor", "Windsurf"]
        license: "MIT"
        note: "Experimental status - breaking changes possible between versions"
        benefit_summary: "AI agents can create actors, build Blueprints, edit node graphs, and control the viewport directly in Unreal Editor."

      unreal_mcp_comprehensive:
        name: "Unreal MCP (ChiR24)"
        repo: "ChiR24/Unreal_mcp"
        search_term: "ChiR24 Unreal_mcp github"
        description: "Comprehensive MCP server with C++ Automation Bridge, TypeScript server, and optional WASM acceleration"
        capabilities:
          - "Manage Blueprints, Materials, Textures, Static Meshes, Skeletal Meshes"
          - "C++ native automation bridge for high-performance operations"
          - "Optional WASM acceleration for compute-intensive tasks"
          - "Optional GraphQL endpoint for complex queries"
          - "Asset type management across the project"
        install_type: "Unreal plugin (C++) + TypeScript MCP server"
        install_steps:
          - "Clone repo and copy plugin to project Plugins folder"
          - "Build the TypeScript MCP server"
          - "Configure MCP client"
        requirements: "Unreal Engine 5.5+"
        supported_clients: ["Claude Desktop", "Cursor", "Windsurf"]
        license: "MIT"
        benefit_summary: "High-performance C++ bridge with broad asset type support and optional WASM acceleration for heavy operations."

  godot:
    engine: "Godot"
    mcps:
      godot_mcp_bradypp:
        name: "Godot MCP (bradypp)"
        repo: "bradypp/godot-mcp"
        search_term: "bradypp godot-mcp github"
        description: "Comprehensive MCP server for Godot with project management, scene editing, and optional read-only mode"
        capabilities:
          - "Launch Godot Editor and execute projects in debug mode"
          - "Generate scenes with specified root node types"
          - "Insert, modify, and remove nodes with customizable properties"
          - "Modify positions, scales, textures, and node attributes"
          - "Auto-load textures into Sprite2D nodes"
          - "Convert 3D scenes to MeshLibrary for GridMap"
          - "Retrieve console output and error messages in real time"
          - "UID management for resource links (Godot 4.4+)"
          - "Read-only mode for safe analysis (CI/CD, code reviews)"
        install_type: "Node.js MCP server (no Godot plugin needed)"
        install_steps:
          - "git clone https://github.com/bradypp/godot-mcp.git"
          - "cd godot-mcp && npm install && npm run build"
          - "Configure MCP client to point to the built index.js"
        requirements: "Godot 3.5+ or 4.0+ (4.4+ for UID features), Node.js"
        supported_clients: ["Claude Desktop", "Cursor", "Cline", "Roo Code"]
        license: "MIT"
        benefit_summary: "Broad Godot version support (3.5+/4.0+) with scene editing, debug output capture, and a read-only mode for safe project analysis."

      godot_mcp_ee0pdt:
        name: "Godot MCP (ee0pdt)"
        repo: "ee0pdt/Godot-MCP"
        search_term: "ee0pdt Godot-MCP github"
        description: "Bidirectional AI bridge for Godot with full project access for scripts, scenes, nodes, and resources"
        capabilities:
          - "Create, delete, and modify nodes in scenes"
          - "Edit and analyze scripts"
          - "Manipulate scenes and project resources"
          - "Access and modify project configuration"
          - "Run projects from the editor"
          - "Two-way communication between AI and Godot"
        install_type: "Node.js MCP server + Godot addon"
        install_steps:
          - "git clone https://github.com/ee0pdt/Godot-MCP.git"
          - "npm install && npm run build"
          - "Copy Godot addon folder to your project's addons directory"
          - "Enable addon in Project > Project Settings > Plugins"
          - "Configure MCP client with server path"
        requirements: "Godot 4.x, Node.js"
        supported_clients: ["Claude Desktop", "Cursor"]
        license: "MIT"
        benefit_summary: "Full bidirectional Godot integration - AI agents can read and write scripts, scenes, and nodes directly."

      godot_mcp_gopeak:
        name: "GoPeak (HaD0Yun)"
        repo: "HaD0Yun/Gopeak-godot-mcp"
        search_term: "Gopeak godot mcp github"
        description: "Comprehensive MCP server for Godot 4.x with ~95+ tools spanning the full edit-run-inspect-fix workflow"
        capabilities:
          - "Find, launch, and manage Godot projects and editor execution"
          - "Create and modify scenes and GDScript files with typed properties"
          - "Work with resources, materials, shaders, and imports"
          - "Debug via logs, LSP diagnostics, DAP breakpoints, and runtime inspection"
          - "Capture screenshots and inspect the live scene tree"
          - "Inject input and call runtime methods for testing"
          - "Discover tools through a catalog with dynamic capability groups"
          - "Access ClassDB introspection and a CC0 asset library"
        install_type: "Node.js MCP server (npx, no Godot plugin needed)"
        install_steps:
          - "Run via 'npx -y gopeak' (or 'npm install -g gopeak')"
          - "Configure your MCP client with paths to the Godot executable and a tool profile"
          - "Choose a tool exposure profile: compact, full, or legacy"
        requirements: "Godot 4.x (4.4+ for UID features), Node.js"
        supported_clients: ["Claude Desktop", "Claude Code", "Cursor"]
        license: "MIT"
        benefit_summary: "The most capable Godot MCP - ~95+ tools, active development, and a one-command npx launch with no Godot plugin required."

      godot_mcp_tugcantopaloglu:
        name: "Godot MCP (tugcantopaloglu)"
        repo: "tugcantopaloglu/godot-mcp"
        search_term: "tugcantopaloglu godot-mcp github"
        description: "MCP server for Godot 4.x with 149 tools, including runtime GDScript execution over a TCP channel"
        capabilities:
          - "Execute arbitrary GDScript at runtime; manipulate nodes and signals"
          - "Control animation, input, raycasting, audio, and video playback"
          - "Headless scene parsing, project-setting edits, and resource creation"
          - "3D/2D rendering, physics joints, skeletal animation, particles, and shaders"
          - "UI theming and networking (HTTP, WebSocket, multiplayer)"
        install_type: "Node.js MCP server + runtime GDScript autoload (no editor plugin)"
        install_steps:
          - "git clone https://github.com/tugcantopaloglu/godot-mcp.git"
          - "cd godot-mcp && npm install && npm run build"
          - "Copy the MCP interaction server GDScript into your project"
          - "Register it as an autoload named 'McpInteractionServer' in Project Settings"
          - "Configure your MCP client with the built server path"
        requirements: "Godot 4.x (4.4+ for UID features), Node.js 18+; TCP port 9090 for runtime tools"
        supported_clients: ["Claude Code", "Cursor", "Cline"]
        license: "MIT"
        benefit_summary: "Largest tool surface (149 tools) with runtime GDScript execution over TCP - strong for live in-game AI interaction and testing."

      godot_mcp_docs:
        name: "Godot MCP Docs"
        repo: "Nihilantropy/godot-mcp-docs"
        search_term: "Nihilantropy godot-mcp-docs github"
        description: "Serves the complete Godot Engine documentation to AI assistants via MCP"
        capabilities:
          - "Query Godot class reference documentation"
          - "Look up tutorials and feature guides"
          - "Search across all Godot documentation"
          - "Retrieve API signatures and usage examples"
        install_type: "Docker container"
        install_steps:
          - "Clone repo and build Docker image"
          - "Configure MCP client to run the containerized server"
        requirements: "Docker, Godot 4.x documentation"
        supported_clients: ["Claude Desktop", "Cursor"]
        license: "MIT"
        benefit_summary: "AI agents get access to the full Godot documentation for accurate API usage without relying on training data."

  roblox:
    engine: "Roblox Studio"
    mcps:
      roblox_studio_mcp:
        name: "Roblox Studio MCP (Official)"
        repo: "Official Roblox Studio plugin — not a community GitHub repo"
        search_term: "Roblox Studio MCP plugin Claude Code setup"
        description: "First-party MCP server built into Roblox Studio. Gives AI agents direct access to the DataModel, scripts, live Luau execution, asset generation, and playtest control — no copy-pasting needed."
        capabilities:
          # DataModel exploration
          - "Investigate the place in parallel and return a compact summary (explore_subagent)"
          - "Explore the instance hierarchy as a flat JSON array (search_game_tree)"
          - "Inspect any Instance's properties, attributes, and children in detail (inspect_instance)"
          # Script tools
          - "Read Scripts, LocalScripts, and ModuleScripts with line numbers (script_read)"
          - "Search all script contents by string or Luau pattern (script_grep)"
          - "Find scripts by name via fuzzy match (script_search)"
          - "Apply multiple edits to a script atomically (multi_edit)"
          # Live execution
          - "Execute arbitrary Luau in the live game context and return the result (execute_luau)"
          - "Retrieve output logs while the game is running (console_output)"
          # Playtest control
          - "Start or stop a playtesting session (start_stop_play)"
          - "Spawn a test character that runs through gameplay scenarios in its own context (playtest_subagent)"
          - "Move the player character to a position or Instance during play (character_navigation)"
          - "Simulate key presses, key holds, and text input (keyboard_input)"
          - "Simulate mouse clicks, movement, and scrolling (mouse_input)"
          # Visual
          - "Capture the current Studio viewport in Play mode (screen_capture)"
          # AI asset generation
          - "Generate a textured 3D mesh from a text prompt (generate_mesh)"
          - "Generate a custom material or texture (generate_material)"
          - "Generate custom procedural models that scale and adapt automatically (generate_procedural_model)"
          - "Insert assets, plugins, and models from the Creator Store (insert_from_creator_store)"
          # Multi-studio
          - "List all connected Studio instances and switch the active one (list_roblox_studios, set_active_studio)"
        install_type: "Roblox Studio Plugin (official, bundled with Studio or available via Plugin Manager)"
        install_steps:
          - "Open Roblox Studio"
          - "Install the MCP plugin from the Studio Plugin Manager if not already present"
          - "The plugin starts a local MCP server when Studio is running"
          - "In Claude Code: claude mcp add roblox-studio <server-command> (verify exact command via Roblox docs)"
          - "Confirm the active Studio instance with list_roblox_studios before making edits"
        requirements: "Roblox Studio (latest version recommended)"
        supported_clients: ["Claude Code", "Claude Desktop"]
        license: "Proprietary (Roblox)"
        note: "Unlike community MCPs for other engines, this is a first-party tool maintained by Roblox. The execute_luau tool — which runs arbitrary Luau in the live game context — is a capability no community engine MCP currently matches. Verify install steps via current Roblox documentation, as the plugin ecosystem evolves."
        benefit_summary: "The official Roblox Studio MCP gives AI agents 21 tools covering the full edit → script → run → inspect → fix loop. The standout capability is execute_luau: run assertions, inspect live state, fire RemoteEvents, and verify game behavior directly from the AI session without leaving Claude Code."

  general:
    engine: "Any"
    description: "General-purpose MCPs useful for any game engine"
    mcps:
      context7:
        name: "Context7"
        repo: "upstash/context7"
        search_term: "upstash context7 MCP github"
        description: "Pulls up-to-date, version-specific documentation and code examples for any library directly into AI prompts"
        capabilities:
          - "Resolve library names to documentation sources"
          - "Retrieve current API documentation for any library or framework"
          - "Pull version-specific code examples and usage patterns"
          - "Verify API signatures against current docs"
        install_type: "npx command or Docker"
        install_steps:
          - "claude mcp add context7 -- npx -y @upstash/context7-mcp"
          - "Or configure in claude_desktop_config.json / .mcp.json"
          - "Optional: get API key at context7.com/dashboard for higher rate limits"
        requirements: "Node.js (for npx) or Docker"
        supported_clients: ["Claude Desktop", "Claude Code", "Cursor", "Windsurf", "Zed"]
        license: "Apache-2.0"
        benefit_summary: "AI agents look up current library documentation instead of relying on training data - reduces outdated API usage across any engine or framework."

# Recommendation logic used by the workflow
recommendation_rules:
  always_recommend:
    - "Engine-specific MCP if one exists for the chosen engine"
    - "Context7 for up-to-date API documentation lookup"

  multiple_options:
    note: "When multiple MCPs exist for an engine, present the top 1-2 options and let the user choose"
    unity_default: "mcp_unity"
    unity_alternative: "unity_mcp_coplay"
    unity_official_note: "unity_mcp_official exists but needs a paid Unity AI subscription - recommend it only when the user already has one"
    unreal_default: "unreal_mcp"
    godot_default: "godot_mcp_gopeak"
    godot_alternative: "godot_mcp_tugcantopaloglu"
    roblox_default: "roblox_studio_mcp"
    roblox_note: "Only one MCP exists for Roblox and it is first-party and official — recommend it unconditionally"

  recommend_if:
    godot_docs_needed:
      trigger: "Godot selected AND beginner game_dev_experience"
      recommend: "godot_mcp_docs for built-in documentation access"

  presentation_order:
    - "engine_specific"
    - "documentation"

# How to present MCPs based on skill level
presentation_guidance:
  expert:
    style: "Concise list with repo links and capabilities"
    example: |
      Available MCPs for {{engine}}:
      - **{{mcp_name}}** ({{repo}}): {{one_line_description}}
        Install: {{install_type}}
      Want to include MCP setup in your architecture? [y/n]

  intermediate:
    style: "Brief explanation with benefits"
    example: |
      **AI Development Tools (MCP Servers)**

      MCP servers let your AI assistant interact directly with {{engine}}.
      For your project, I'd recommend:

      - **{{mcp_name}}**: {{benefit_summary}}
        Repo: {{repo}}

      These are optional but significantly improve AI-assisted development.
      Include MCP setup in your architecture? [y/n]

  beginner:
    style: "Full explanation of what MCPs are and why they help"
    example: |
      **AI Development Tools**

      There's a way to give your AI assistant "eyes" into {{engine}} -
      called MCP (Model Context Protocol) servers. Think of it like giving
      your AI assistant a window into your game editor.

      For {{engine}}, there's **{{mcp_name}}** which lets the AI:
      {{friendly_capability_list}}

      This means instead of describing your scene layout in chat,
      the AI can just look at it directly.

      Want to set this up? [y/n]