DEV Community: Thomas Gorisse

6 MCP Servers for 3D & AR Development — What AI Can Build Now

Thomas Gorisse — Mon, 13 Apr 2026 00:14:41 +0000

What if you could ask Claude "build me a car configurator" and get compilable Android code in 10 seconds?

MCP (Model Context Protocol) makes this real. An MCP server gives AI assistants access to specialized tools — code generators, validators, domain knowledge — so they produce working code instead of generic suggestions.

I built 6 MCP servers for 3D and AR development. Here's what each one does and how to use them.

What Is MCP?

MCP is an open protocol (by Anthropic) that lets AI assistants call external tools. Think of it as plugins, but standardized — one server works with Claude, Cursor, Windsurf, and any MCP-compatible client.

An MCP server provides:

Tools: functions the AI can call (generate code, validate, search)
Resources: data the AI can read (API docs, samples)
Prompts: pre-built workflows (code review, scaffolding)

The 6 Servers

1. sceneview-mcp — General 3D & AR (28 tools)

The flagship. Covers all SceneView features across Android, iOS, and Web.

npx sceneview-mcp

Key tools:

generate_scene — complete scene code from a description
validate_code — catches threading bugs, null handling, wrong APIs
search_models — searches Sketchfab for free 3D models
get_sample — returns working sample code for any feature

Example prompt: "Create a 3D product viewer with orbit camera and environment lighting"

2. automotive-3d-mcp — Car Configurators & HUD

npx automotive-3d-mcp

Specialized for automotive UX: car configurators, HUD overlays, 3D dashboards, AR showrooms, parts catalogs. Knows ADAS display patterns, Tier-1 integration requirements, and automotive UX standards.

Example prompt: "Build a car color configurator with paint + wheel options"

3. healthcare-3d-mcp — Medical Visualization

npx healthcare-3d-mcp

Anatomy viewers, DICOM slice rendering, molecular structures, surgical planning tools, dental scanning visualization. Includes HIPAA-compliant patterns and medical imaging standards.

Example prompt: "Create an interactive anatomy model with organ highlighting"

4. gaming-3d-mcp — Game Development

npx gaming-3d-mcp

Character controllers, terrain generation, particle systems, physics interactions, inventory UI, NPC behaviors. Covers casual games, not AAA — optimized for mobile.

Example prompt: "Generate a third-person character controller with jump and attack"

5. interior-design-3d-mcp — Room Planning

npx interior-design-3d-mcp

Room planners, AR furniture placement, material/finish switching, lighting simulation, measurement tools. Knows furniture catalog patterns and real estate staging.

Example prompt: "Build an AR room planner where users tap to place furniture"

6. rerun-3d-mcp — AR Debug Visualization

npx rerun-3d-mcp

Integration with Rerun.io for AR debugging. Stream point clouds, camera poses, anchor positions, and depth maps from your AR app to Rerun's web viewer in real-time.

Example prompt: "Set up Rerun bridge to visualize my AR app's point cloud"

How to Install

Claude Code (CLI):

claude mcp add sceneview -- npx sceneview-mcp

Claude Desktop / Cursor / Windsurf:

{
  "mcpServers": {
    "sceneview": {
      "command": "npx",
      "args": ["-y", "sceneview-mcp"]
    }
  }
}

Replace sceneview-mcp with any of the 6 server names above.

What Makes a Good 3D MCP Server

After building 6 servers, here's what matters:

Complete, compilable code — not snippets. The output must build and run.
Validation tools — catch common mistakes (wrong thread, null model, missing permissions) before the developer runs the code.
Domain knowledge — not just API docs, but patterns. "A car configurator needs X, Y, Z" is more useful than "here's the ModelNode API."
Working samples — every tool returns code that was tested against the real SDK.

Try It Now

The fastest way to test:

# Install Claude Code if you haven't
npm install -g @anthropic-ai/claude-code

# Add SceneView MCP
claude mcp add sceneview -- npx sceneview-mcp

# Ask it to build something
claude "Build me an AR app that places 3D furniture on detected surfaces"

All servers are open source and available on npm:

GitHub: github.com/sceneview/sceneview

All SceneView MCP servers are open source (Apache 2.0). Contributions welcome.

The Only Compose-Native 3D & AR Library for Android

Thomas Gorisse — Mon, 13 Apr 2026 00:07:17 +0000

You're building an Android app with Jetpack Compose. You need 3D. Maybe a product viewer, a model configurator, or AR furniture placement.

You search. You ask ChatGPT. You try Copilot. And you get: Unity (50MB+, no Compose), raw Filament (1000 lines to render a cube), or Sceneform (deprecated since 2021).

There's a gap. Android has no built-in 3D composable. Until now.

SceneView: 3D as a Composable

SceneView treats 3D the same way Compose treats UI: declaratively. SceneView { } is a composable like Column { } or LazyList { }. Nodes are composables. State is Kotlin state.

implementation("io.github.sceneview:sceneview:4.0.0")

Display a 3D model in 15 lines

@Composable
fun ProductViewer() {
    val engine = rememberEngine()
    val modelLoader = rememberModelLoader(engine)

    SceneView(
        modifier = Modifier.fillMaxSize(),
        engine = engine,
        modelLoader = modelLoader,
        cameraManipulator = rememberCameraManipulator()
    ) {
        rememberModelInstance(modelLoader, "models/sneaker.glb")?.let {
            ModelNode(modelInstance = it, scaleToUnits = 1.0f, autoAnimate = true)
        }
    }
}

Place sneaker.glb in src/main/assets/models/. The model loads asynchronously, rememberModelInstance returns null while loading, then triggers recomposition. No callbacks, no lifecycle — just state.

What you get out of the box

SceneView includes 26+ composable node types:

ModelNode — GLB/glTF models with animation
CubeNode, SphereNode, CylinderNode — primitive geometry
ImageNode — textured quads
VideoNode — video textures on 3D surfaces
LightNode — point, spot, directional lights
TextNode — 3D text billboards
ViewNode — any Compose UI in 3D space

All powered by Google Filament — the same PBR engine used in Google Maps and Android Auto.

AR in Compose: One More Import

implementation("io.github.sceneview:arsceneview:4.0.0")

@Composable
fun ARFurniturePlacement() {
    val engine = rememberEngine()
    val modelLoader = rememberModelLoader(engine)
    val anchors = remember { mutableStateListOf<Anchor>() }
    val chair = rememberModelInstance(modelLoader, "models/chair.glb")

    ARSceneView(
        modifier = Modifier.fillMaxSize(),
        engine = engine,
        modelLoader = modelLoader,
        planeRenderer = true,
        onTouchEvent = { hitResult, _ ->
            hitResult?.let { anchors.add(it.createAnchor()) }
        }
    ) {
        anchors.forEach { anchor ->
            chair?.let { model ->
                AnchorNode(anchor = anchor) {
                    ModelNode(
                        modelInstance = model,
                        scaleToUnits = 0.8f,
                        isEditable = true
                    )
                }
            }
        }
    }
}

isEditable = true gives you drag, pinch-to-scale, and two-finger rotate. AR state is Compose state — when anchors changes, the scene recomposes.

Supported AR features: plane detection, image tracking, face mesh, cloud anchors, geospatial anchors, depth occlusion, light estimation.

Why Not Unity / Filament / Sceneform?

	SceneView	Unity	Raw Filament	Sceneform
Compose native	Yes	No	No	No
APK size	~5MB	~50MB+	~3MB	~5MB
AR built-in	Yes (ARCore)	Plugin	No	Yes
Active	v4.0.0 (2026)	Yes	Yes	Deprecated (2021)
Learning curve	Low	High	Very high	Low
Scene graph	26+ node types	Full engine	None	5 node types
Cross-platform	Android, iOS, Web	All	Android only	Android only

SceneView wraps Filament internally — same rendering quality, 100x less code.

Cross-Platform: Same API, Native Renderers

SceneView also supports:

iOS/macOS/visionOS: SwiftUI + RealityKit (native Apple renderer)
Web: Kotlin/JS + Filament.js (WASM)
Flutter: PlatformView bridge
React Native: Fabric bridge

Each platform uses its native renderer — no cross-compiled runtime. The API mirrors across platforms: SceneView { ModelNode(...) } works on Android and iOS.

AI-First: Ask Any AI to Build Your 3D App

SceneView is designed so that AI assistants generate correct code on the first try:

llms.txt — 3000-line machine-readable API reference
MCP Server — 28 tools for Claude/Cursor (install: npx sceneview-mcp)
Copilot/Cursor/Windsurf rules — IDE-specific code generation patterns

Try it: open Claude and ask "build me an AR app with SceneView." The MCP server provides validated samples and the full API, so you get compilable code.

Industry-specific MCP servers are also available:

npx automotive-3d-mcp — car configurators
npx healthcare-3d-mcp — medical visualization
npx gaming-3d-mcp — game development
npx interior-design-3d-mcp — interior design

Get Started

// build.gradle.kts
implementation("io.github.sceneview:sceneview:4.0.0")    // 3D only
implementation("io.github.sceneview:arsceneview:4.0.0")   // 3D + AR

GitHub (1.2k+ stars)
Full API Reference
MCP Server (13k+ monthly installs)
Discord

SceneView is open source (Apache 2.0) and actively maintained. Try it in your next Compose project.