Team Development with Game Studios
Claude Code Game Studios is a project that streamlines game development with 48 specialized AI agents and 37 slash commands. When integrated with OpenForge MCP, each agent can directly operate Unity / Blender / Godot to perform level construction, art adjustments, QA testing, and more.
This tutorial covers setup, workflow execution, and customization.
Prerequisites
- OpenForge MCP is installed and connected to a game engine
- You understand the basic Game Studios concepts (see Game Studios Integration)
What Is Game Studios?
Game Studios defines each game development role as an AI agent. Each agent is assigned tool permissions corresponding to their area of expertise.
creative-director --- Oversees vision and quality
technical-director -- Technical decision-making
level-designer ------ Scenes, terrain, NavMesh
art-director -------- Lighting, materials, VFX
technical-artist ---- Optimization, shaders
gameplay-programmer - Physics, input, animation, scripting
ai-programmer ------- NavMesh, ML-Agents
qa-tester ----------- Playtesting, profiling
qa-lead ------------- Bug reports, quality management
3d-modeler ---------- Blender modeling
animator ------------ Animation, timeline
sound-designer ------ Audio
...
Step 1: Setup
Clone the Game Studios Repository
git clone https://github.com/Donchitos/Claude-Code-Game-Studios.git
cd Claude-Code-Game-Studios
Configure Integration with OpenForge MCP
You:
Integrate the Game Studios project with OpenForge.
AI Response Example:
Executed setup_game_studios.
Generated configuration files:
.claude/settings.json
- Added OpenForge MCP server settings
- Configured tool permissions for 22 agent roles
Connection status:
Unity: Connected (OpenForge MCP v0.8.0)
Blender: Connected (OpenForge MCP v0.8.0)
Godot: Not connected
Available agents: 22
Available workflows: 8
Check an Agent's Tool Permissions
You:
Show me the tools available to the level-designer.
AI Response Example:
level-designer tool permissions:
Categories: Scene, GameObject, Terrain, Lighting, NavMesh, Prefab, Weather, Camera
Key tools (42 total):
Scene: create_scene, save_scene, load_scene
GameObject: create_gameobject, set_transform, duplicate_gameobject, delete_gameobject, ...
Terrain: create_terrain, set_terrain_height, paint_terrain, set_terrain_trees, ...
Lighting: create_light, set_light_property (placement only; adjustments are handled by art-director)
NavMesh: bake_navmesh, set_navmesh_area, create_navmesh_obstacle, ...
Prefab: instantiate_prefab, create_prefab
Weather: set_fog, set_skybox
Camera: set_camera, take_screenshot
Step 2: Run the "Create Level" Workflow
Predefined workflows have multiple agents work through tasks sequentially.
You:
Run the create_level workflow. Create a "medieval castle courtyard". Size 50m x 50m. Cobblestone floor, fountain in the center, torches at the four corners.
AI Response Example:
Starting create_level workflow.
=== Step 1/3: level-designer ===
Tools used: create_scene, create_terrain, create_gameobject, set_transform
Running...
- Created scene "CastleCourtyard"
- Placed ground (50m x 50m)
- Placed fountain base (center) -- composed of Cylinder + Torus
- Placed 4 torch stands at corners
- Built castle walls (perimeter) with Cubes
- Placed gate (front entrance)
- Placed 4 benches along the pathways
- Baked NavMesh
level-designer complete. Object count: 28
=== Step 2/3: art-director ===
Tools used: create_material, create_light, set_skybox
Running...
- Created M_Cobblestone (cobblestone material), applied to floor
- Created M_CastleWall (castle wall material), applied to walls
- Created M_Water (water material), applied to fountain
- Set directional light (sunset color)
- Placed 4 point lights at torch positions (warm color, with flicker script)
- Set skybox to overcast
- Captured screenshot
art-director complete. Materials: 5, Lights: 5
=== Step 3/3: technical-artist ===
Tools used: analyze_draw_calls, create_lod_group, optimize_textures
Running...
- Draw call analysis: 34 (good)
- Combined castle wall meshes (12 -> 1 draw call)
- Applied texture compression
- Final draw calls: 23
technical-artist complete.
=== Workflow Complete ===
Scene: CastleCourtyard
Object count: 28
Draw calls: 23
Screenshot saved.
Step 3: Run the "QA Pass" Workflow
Automatically check the quality of the constructed level.
You:
Run the QA Pass workflow. Check the CastleCourtyard scene.
AI Response Example:
Starting qa_pass workflow.
=== Step 1/3: qa-tester (Playtest) ===
Tools used: enter_play_mode, simulate_input, take_screenshot
Running...
- Entered play mode
- Spawned player at spawn point
- Auto-movement test: circled around the fountain
- Collision test: wall collision detection
- NavMesh test: reachability to all areas
Screenshots: 8 captured
qa-tester (Playtest) complete.
=== Step 2/3: qa-tester (Bug Recording) ===
Tools used: take_screenshot, exit_play_mode
Issues found:
[BUG-001] Severity: Medium
NavMesh gap on the south side of the fountain. NPCs cannot traverse.
Screenshot: qa_bug001.png
[BUG-002] Severity: Low
Bench #3 BoxCollider is larger than the mesh.
Player hits an invisible wall.
[BUG-003] Severity: Info
Torch #2 light bleeds through the wall. Range should be adjusted.
qa-tester (Bug Recording) complete. Bugs: 3
=== Step 3/3: qa-lead (Report) ===
Tools used: take_screenshot
QA Report:
Test items: 12
Passed: 9
Failed: 2
Info: 1
Recommended actions:
1. Re-bake NavMesh (fix BUG-001)
2. Adjust Bench #3 BoxCollider size (fix BUG-002)
3. Change Torch #2 Light Range from 8 to 5 (address BUG-003)
qa-lead complete.
=== Workflow Complete ===
You:
Fix all the bugs from the QA report.
AI Response Example:
Executing bug fixes.
[BUG-001] Fixed NavMesh gap
- Found a gap in the ground mesh on the south side of the fountain
- Filled the gap and re-baked NavMesh
- Verified: NPC pathfinding succeeds for all points
[BUG-002] Adjusted BoxCollider size
- Bench_3 BoxCollider: (2.5, 1.0, 1.0) -> (1.8, 0.8, 0.6)
- Applied auto-fit
[BUG-003] Adjusted Light Range
- TorchLight_2: Range 8 -> 5
- Wall bleed-through resolved
All 3 fixes complete.
Step 4: Customize Workflows
You can modify predefined workflows to create project-specific flows.
Workflow Definition File
Workflows are defined in YAML. Here is an example of a custom "create_dungeon" workflow.
name: create_dungeon
description: Auto-generate a dungeon
version: "1.0"
steps:
- agent: level-designer
tools:
- create_scene
- create_gameobject
- set_transform
- duplicate_gameobject
prompt: |
Build the basic dungeon structure.
- Generate 5-8 rooms (random sizes: 5-15m)
- Connect rooms with corridors
- Place entrance and exit
- Place PropSpawnPoint (Empty) in each room
- agent: art-director
tools:
- create_material
- set_material
- set_material_property
- create_light
- set_light_property
prompt: |
Set up the dungeon visuals.
- Create stone wall material and apply to walls
- Cracked stone material for floors
- Point light in each room (dim warm color)
- Torch lights in corridors (with flicker)
- Overall dark atmosphere
- agent: ai-programmer
tools:
- bake_navmesh
- set_navmesh_area
- create_navmesh_obstacle
prompt: |
Set up NavMesh for NPCs.
- All corridors and rooms must be connected via NavMesh
- Place NavMesh Obstacles at door positions (for opening/closing)
- Set NavMesh Area to "Narrow" for tight corridors
- agent: gameplay-programmer
tools:
- create_script
- add_component
- set_component_property
prompt: |
Create dungeon gameplay logic.
- DungeonManager script: room clear state management
- DoorController script: door opens on approach
- TrapTrigger script: trap activation logic
- Apply scripts to PropSpawnPoints
- agent: qa-tester
tools:
- enter_play_mode
- simulate_input
- take_screenshot
- exit_play_mode
prompt: |
Playtest the dungeon.
- Can the player reach the exit from the entrance?
- Can all rooms be entered?
- Do doors open and close correctly?
- Is the lighting too dark? (verify with screenshots)
- agent: technical-artist
tools:
- analyze_draw_calls
- optimize_textures
- create_lod_group
- combine_meshes
prompt: |
Performance optimization.
- Keep draw calls under 200
- Combine static meshes
- Apply texture compression
Running the Workflow
You:
Run the create_dungeon workflow. Theme: "catacombs".
AI Response Example:
Starting create_dungeon workflow. Theme: Catacombs
=== Step 1/6: level-designer ===
- Created scene "Catacomb_01"
- Generated 6 rooms (6m-12m)
- Connected with 8 corridors
- Entrance: north end / Exit: deepest room at south end
- PropSpawnPoints: 18 placed
Object count: 94
=== Step 2/6: art-director ===
- M_CatacombWall: dark gray stone wall with normal map
- M_CatacombFloor: cracked stone
- M_Bones: bone material (for PropSpawnPoints)
- Point lights: 1-2 per room (eerie pale green)
- Corridors: torch lights (warm, flickering)
Materials: 4, Lights: 14
=== Step 3/6: ai-programmer ===
- NavMesh bake complete
- NavMesh Obstacles: 8 doors
- Set 3 narrow corridors to "Narrow" area
- Path test: entrance -> exit reachable
=== Step 4/6: gameplay-programmer ===
- DungeonManager.cs created
- DoorController.cs created (proximity trigger)
- TrapTrigger.cs created (floor trap)
- Scripts applied to 18 PropSpawnPoints
=== Step 5/6: qa-tester ===
- Playtest run (30 seconds)
- Entrance -> exit: reachable
- All 6 rooms: accessible
- Doors: 8/8 functioning correctly
- Bugs: 1 (room 4 trap does not activate -- TriggerCollider isTrigger is false)
Screenshots: 12
=== Step 6/6: technical-artist ===
- Draw calls: 156 (target under 200: achieved)
- Combined wall meshes per room (48 -> 6)
- Applied texture compression
=== Workflow Complete ===
Auto-fixed 1 bug: set isTrigger = true on room 4 TrapTrigger
Step 5: Add Custom Agent Definitions
You can add project-specific specialist agents.
Agent Definition File
# agents/vr-specialist.yaml
name: vr-specialist
description: VR experience specialist. Handles interaction design and comfort
tools:
- category: Input
- category: Physics
- category: Camera
- category: UI
- category: Script
- tool: take_screenshot
- tool: enter_play_mode
- tool: exit_play_mode
prompts:
default: |
Optimize VR experience quality. Priorities:
- Hand tracking interaction design
- Motion sickness prevention (field of view restriction, teleport locomotion)
- UI placement (world space UI, gaze-following)
- Performance (maintain 90+ FPS)
Registering the Agent
You:
Register the vr-specialist agent.
AI Response Example:
Agent "vr-specialist" registered.
Available tools: 38
Input: 8 tools
Physics: 12 tools
Camera: 4 tools
UI: 6 tools
Script: 5 tools
Individual: take_screenshot, enter_play_mode, exit_play_mode
To use in a workflow:
steps:
- agent: vr-specialist
prompt: "Design VR interactions"
Incorporating Custom Agents into Workflows
name: create_vr_level
description: Build a VR-ready level
version: "1.0"
steps:
- agent: level-designer
prompt: "Build level geometry"
- agent: art-director
prompt: "Set up visuals"
- agent: vr-specialist
tools:
- create_script
- add_component
- set_component_property
- enter_play_mode
- take_screenshot
prompt: |
Set up VR interactions.
- Add XR Grab Interactable to grabbable objects
- Define teleport areas
- Place UI in world space (1.5m in front of the player)
- Verify 90+ FPS in playtest
- agent: qa-tester
prompt: "Playtest and record bugs"
Agent Roles and Tool Permissions Reference
Common agents and their permission mappings.
| Agent | Tool Categories | Primary Use |
|---|---|---|
| level-designer | Scene, GameObject, Terrain, NavMesh, Prefab | Level layout |
| art-director | Material, Lighting, VFX, Camera | Visual quality |
| technical-artist | Material, Optimization, Mesh, Bake, UV | Shaders, optimization |
| gameplay-programmer | Physics, Input, Animation, Script | Game logic |
| ai-programmer | NavMesh, ML-Agents, Script | NPC AI |
| qa-tester | Playtest, Camera, Optimization | Test execution |
| qa-lead | Playtest, Camera | Report creation |
| 3d-modeler | Blender Object/Mesh/Procedural/Modifier | Modeling |
| animator | Animation, Timeline, Blender Animation | Animation |
| sound-designer | Audio | Sound |
Workflow design tip: Design agent responsibilities so they do not overlap. When multiple agents operate on the same object, use step ordering to prevent conflicts.