How do I add 3D to my Android Jetpack Compose app?

Use SceneView SDK — the only Compose-native 3D library for Android. Add implementation("io.github.sceneview:sceneview:4.37.0") to your build.gradle, then use SceneView { ModelNode(...) } composable. SceneView wraps Google Filament with a declarative API, supports 48+ node types, and adds only ~5MB to your APK.

What is the best AR SDK for Android with Jetpack Compose?

SceneView's ARSceneView is the only Compose-native AR SDK for Android. It wraps ARCore with a declarative API: add implementation("io.github.sceneview:arsceneview:4.37.0"), then use ARSceneView { AnchorNode(...) } composable. Supports plane detection, image tracking, cloud anchors, geospatial anchors, and depth.

How do I build a 3D app for iOS with SwiftUI?

Use SceneViewSwift via Swift Package Manager. Add https://github.com/sceneview/sceneview.git as a dependency. It provides SceneView and ARSceneView SwiftUI views powered by RealityKit, supporting iOS, macOS, and visionOS.

Is there an MCP server for 3D and AR development?

Yes — sceneview-mcp is a Model Context Protocol server for AI-assisted 3D and AR development. Install with npx -y sceneview-mcp. It provides 32 tools including code generation, validation, samples, and guides for Android, iOS, and Web. Works with Claude Code, Cline, Codex, Cursor, GitHub Copilot, JetBrains AI Assistant and any other MCP-compatible client.

How do I open a .3mf file on Android?

SceneView reads 3MF — the format ChatGPT, slicers and image-to-print flows emit. There is no separate API: ModelLoader detects the payload by its ZIP magic and converts it to GLB in memory, so rememberModelInstance(modelLoader, uri) opens a .3mf exactly like a .glb, in 3D and in AR. Millimetres are scaled to metres and the printer's Z-up is rotated to Y-up, so a 60 mm print is life-size and upright.

Is there a SceneView plugin for ChatGPT or Codex?

Yes — the SceneView repository is also a Codex plugin. Its local marketplace catalog is .agents/plugins/marketplace.json (.codex-plugin/plugin.json is kept as the compatibility fallback), and it points at three skills (Jetpack Compose, SwiftUI, web) carrying the full API contract. From a checkout, run codex plugin marketplace add . — see the Codex documentation for the current install flow. For ChatGPT, npx -y sceneview-mcp --http serves the MCP Streamable HTTP transport plus an inline 3D model viewer widget.

What is the difference between SceneView and Unity/Unreal for mobile 3D?

SceneView is purpose-built for mobile apps (not games). It adds ~5MB vs 50-100MB+ for Unity/Unreal, uses native UI frameworks (Jetpack Compose, SwiftUI), and integrates with the Android/iOS ecosystem. Unity/Unreal are better for complex games; SceneView is better for 3D features in regular apps (product viewers, AR, data visualization).

The 3D & AR SDK that AI gets right on the first try

Name: SceneView
Author: SceneView

SceneView is designed so any AI coding assistant can produce correct 3D & AR code on the first try — for Jetpack Compose, SwiftUI, the Web, and beyond. It opens glTF/GLB, 3MF, STL, OBJ and PLY on Android and USDZ on Apple, and shows them at real size in AR.

Get Started View on GitHub

Try the demo:

Web Playground

Android

iOS

Web

visionOS

macOS

Flutter

React Native

3D Scene Preview

Loading 3D engine…

Rendered live by SceneView Web (Filament.js WASM)

Platforms

26+

3D node types

Lines for 3D

60fps

PBR on mobile

MCP server

~3MB

APK overhead

Get started in seconds

Pick your platform, paste one line, and you are running 3D.

// build.gradle.kts
dependencies {
    // 3D only
    implementation("io.github.sceneview:sceneview:4.37.0")
    // 3D + AR
    implementation("io.github.sceneview:arsceneview:4.37.0")
}

// Swift Package Manager
// Add in Xcode: File → Add Package Dependencies

"https://github.com/sceneview/sceneview"
// Version: 4.37.0

import SceneViewSwift

// One line in your HTML:
<script src="https://cdn.jsdelivr.net/npm/sceneview-web@4/sceneview-web.js"></script>

// Then use the API:
SceneView.modelViewer("canvas", "model.glb")

# pubspec.yaml (coming soon on pub.dev)
dependencies:
  sceneview_flutter:
    git:
      url: https://github.com/sceneview/sceneview
      path: flutter

// package.json
"dependencies": {
  "react-native-sceneview": "github:sceneview/sceneview#react-native"
}

// Then run:
npm install
npx pod-install // iOS only

// Compose Desktop: wireframe placeholder only
// (no 3D renderer yet — Filament JVM pending)
git clone https://github.com/sceneview/sceneview
cd sceneview/samples/desktop-demo
./gradlew run

AI-first SDK

Built for AI assistants

One MCP server, and your assistant stops guessing at the SceneView API. It compile-checks the Kotlin or Swift it just wrote, looks up the exact node API instead of inventing one, and starts you from a sample that already builds — plus AR setup, gestures, and troubleshooting when you get stuck. The repository ships as a ChatGPT and Codex plugin too, so the same API contract travels wherever you work.

Claude Code Cline Codex Cursor Gemini in Android Studio GitHub Copilot JetBrains AI Assistant Any MCP client

auto_awesome Set up your assistant package_2 View on npm

AI assistant — SceneView Project

You Build me a 3D product viewer with AR try-on for shoes

Assistant

Using SceneView MCP to generate your app…

// MainActivity.kt

ARScene(

planeRenderer = true,

onTapAR = { hitResult ->

AnchorNode(hitResult) {

ModelNode("shoe.glb")

}

)

✓ Compiles on first try. Tap the floor to place the shoe.

You Add a hotspot on the laces that opens a size selector

Assistant

Reading SceneView hotspot & gesture docs via MCP…

✓ Hotspot added. Preview updated in Android emulator.

Use SceneView with your AI coding assistant

One MCP server, every assistant. Pick yours, copy one line, and it can read the full SceneView API reference, the samples and the best practices before it writes a line of 3D code.

Locally, over stdio

npx -y sceneview-mcp

Remotely, over Streamable HTTP — already hosted, nothing to run

https://mcp.sceneview.dev/mcp

Config shapes differ across the ecosystem, the command and its arguments never do. The common one is mcpServers; VS Code uses servers with "type": "stdio", Codex a [mcp_servers.*] TOML table, Copilot CLI "type": "local", and Kilo Code a mcp key in kilo.jsonc where command is an array and the environment block is called environment. When in doubt, paste the command above into whatever field your client offers.

Run in your terminal

claude mcp add sceneview -- npx -y sceneview-mcp

Or commit it for the whole team — .mcp.json

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

Reference: code.claude.com/docs/en/mcp

MCP Servers icon → Configure → Configure MCP Servers — or ~/.cline/mcp.json

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

Reference: docs.cline.bot/mcp/configuring-mcp-servers

Run in your terminal

codex mcp add sceneview -- npx -y sceneview-mcp

Or edit ~/.codex/config.toml — same config for the CLI, the IDE extension and the app

[mcp_servers.sceneview]
command = "npx"
args = ["-y", "sceneview-mcp"]

Reference: learn.chatgpt.com/docs/extend/mcp

Add to .cursor/mcp.json (project) or ~/.cursor/mcp.json (global)

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

Cursor also accepts an install link: open in Cursor (requires Cursor installed). Reference: cursor.com/docs/mcp/install-links

Settings → Tools → AI → MCP Servers — Android Studio connects over HTTP, not stdio

{ "mcpServers": { "sceneview": {
  "httpUrl": "https://mcp.sceneview.dev/mcp",
  "enabled": true
} } }

Gemini in Android Studio accepts httpUrl only — the local npx -y sceneview-mcp stdio command cannot be registered here, so this entry points at the hosted server instead. If you would rather run it locally inside the same IDE, use Junie: put the Any MCP client snippet in .junie/mcp/mcp.json. Reference: developer.android.com/studio/gemini/add-mcp-server

Add to .vscode/mcp.json — note the servers key, not mcpServers

{ "servers": { "sceneview": {
  "type": "stdio",
  "command": "npx", "args": ["-y", "sceneview-mcp"]
} } }

Copilot CLI

copilot mcp add sceneview -- npx -y sceneview-mcp

Writing the CLI's file by hand takes a third shape: ~/.copilot/mcp-config.json uses mcpServers like most clients, but each server needs "type": "local" instead of "stdio". VS Code also accepts an install link: open in VS Code (requires VS Code installed). References: code.visualstudio.com/docs/agents/reference/mcp-configuration and docs.github.com — add MCP servers to Copilot CLI

Settings → Tools → AI Assistant → Model Context Protocol (MCP) → Add, then paste

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

Works for AI Assistant and Junie, in IntelliJ IDEA, Android Studio and the other JetBrains IDEs. Junie also reads the same JSON straight from a file — .junie/mcp/mcp.json for one project, ~/.junie/mcp/mcp.json for every project. Reference: jetbrains.com/help/ai-assistant/mcp.html

Agents running inside Xcode read the rules files at your project root

# at your project root
AGENTS.md
CLAUDE.md

Apple recommends putting your project guidance in these files, and the agents running inside Xcode read them. Xcode 26.3 itself exposes its own tools to external agents rather than hosting this server: turn it on under Settings → Intelligence → Model Context Protocol, then run your CLI alongside Xcode with both servers — claude mcp add --transport stdio xcode -- xcrun mcpbridge or codex mcp add xcode -- xcrun mcpbridge — plus SceneView from the Claude Code or Codex tab. Reference: developer.apple.com — external agents.

No MCP support? Point your assistant at the plain-text reference

https://sceneview.github.io/llms.txt

It is the whole SDK — API surface, samples and best practices — in one file any assistant can read. The repository also ships AGENTS.md, CLAUDE.md, .github/copilot-instructions.md and .cursorrules, so whichever rules file your tool reads, it finds the same guidance. Full docs: SceneView documentation.

Also works with Antigravity, Continue, Devin Desktop, Goose, Kilo Code, Kiro, OpenCode, OpenHands, Qwen Code, Windsurf, Zed — and any MCP-compatible client.

Scene(
    modifier = Modifier.fillMaxSize(),
    nodes = listOf(
        ModelNode(
            model = "models/drone.glb",
            autoAnimate = true,
            scale = 0.5f
        )
    ),
    environment = Environment(
        hdr = "skybox.hdr"
    )
)

SceneView {
    Model(named: "drone")
        .rotation(x: .pi/4)
        .scale(0.5)
        .onTap {
            print("Drone selected")
        }
}
.environment(named: "outdoor")
.edgesIgnoringSafeArea(.all)

// One line in your HTML:
<script src="sceneview.js"></script>

// Render a 3D model:
SceneView.modelViewer("canvas", "model.glb", {
  backgroundColor: [0.05, 0.05, 0.08, 1],
  lightIntensity: 150000,
  fov: 35
})

// Ask your AI assistant, with SceneView MCP:

"Build me an AR app that lets users
 place 3D furniture in their room"

// It generates correct, working SceneView
// code on the first try.

// Setup, in any MCP client:
npx -y sceneview-mcp

Stable

Android

Jetpack Compose + Filament + ARCore

Filament (Vulkan/OpenGL) Alpha

iOS

SwiftUI + RealityKit + ARKit

RealityKit (Metal) Alpha

macOS

SwiftUI + RealityKit

RealityKit (Metal) Alpha

visionOS

SwiftUI + RealityKit Spatial

RealityKit (Metal) Stable

Web

Kotlin/JS + Filament.js WASM

Filament (WebGL2) Alpha

Android TV

Compose TV + D-pad controls

Filament (Vulkan) Not yet

Desktop

Compose Desktop · wireframe placeholder

No 3D renderer yet (Filament JVM pending)

Coming Soon

Android XR

Jetpack XR SceneCore

Compose XR

Alpha

Flutter

PlatformView bridge

Native (Android + iOS) Alpha

React Native

Fabric bridge

Native (Android + iOS)

What you can build

48+ node types, full ARCore + ARKit coverage, physics, post-processing, and Compose UI inside a 3D scene.

view_in_ar

46+ Node Types

Models, primitives (cube, sphere, cone, torus, capsule), curves, shapes, text, video, billboard sprites, lights, reflection probes, dynamic sky, fog, physics, and more — every one a Composable / SwiftUI view.

Android iOS Web

layers

Compose UI in 3D

ViewNode renders any @Composable as a textured plane in world space. Buttons, lists, animations — all interactive, all reactive. The same trick exists in SwiftUI on iOS.

Android iOS

public

AR Geospatial

Streetscape city mesh, terrain anchors, rooftop anchors, Cloud Anchors for cross-device persistence. Build navigation, geofenced experiences, and scavenger hunts at planet scale.

Android ARCore VPS

face

Augmented Faces & Images

Face mesh overlay (front camera) with custom materials. Image tracking for posters, business cards, museum exhibits — with runtime image database building.

Android iOS

videocam

AR Record & Replay

Capture an outdoor ARCore session once with ARRecorder, replay it 1:1 at the desk via ARSceneView(playbackDataset = file). Debug AR scenes without a phone in your hand.

Android EIS — image stabilization

animation

Physics & Gestures

Rigid-body physics (gravity, collisions, impulses), ray vs box/sphere collision, drag-rotate-scale-elevate gestures — all opt-in per node via isEditable.

All platforms

brightness_high

PBR Lighting & Post-FX

Image-based lighting from .hdr / .ktx, dynamic time-of-day sky, atmospheric fog, reflection probes, plus bloom, depth of field, SSAO, vignette, color grading, and tone mapping.

Filament RealityKit

vrpano

WebXR — AR & VR

Same engine on the web. SceneView.startAR(canvas) for phone passthrough AR with hit-test, anchors, light estimation. startVR(canvas) for headset VR with controllers and reference spaces.

Web immersive-ar immersive-vr

smart_toy

AI-First MCP Server

validate_code compile-checks what your assistant wrote, get_node_reference resolves the exact node API, get_setup and get_ar_setup wire up the project, and list_samples starts you from code that builds. 38 compilable samples + full llms.txt reference.

Claude Code Cline Codex Cursor Gemini in Android Studio GitHub Copilot JetBrains AI Assistant Any MCP client

deployed_code

Opens the `.3mf` AI makes

Ask a chat assistant for a 3D print and it hands back a .3mf — a file nothing else on Android views in 3D, let alone in AR. SceneView reads it with no new API: the payload is detected by its ZIP magic and converted to GLB in memory, scaled from millimetres to metres and rotated upright, so a 60 mm print is life-size on a real table. The Android demo registers for the type, so a shared file opens straight from the chooser.

3MF glTF / GLB USDZ

How SceneView compares

The lightweight alternative to heavy game engines for 3D in native apps.

	SceneView	Unity	Filament (direct)	Three.js	Babylon.js	RealityKit SDK
Setup time	2 min (1 dep)	30+ min	15 min (C++ build)	5 min (npm)	5 min (npm)	10 min (Xcode)
Binary overhead	~3 MB	~30 MB+	~2 MB	~600 KB (JS)	~800 KB (JS)	Built-in
Compose / SwiftUI	Native (first-class)	None	None (imperative C++)	N/A (web only)	N/A (web only)	SwiftUI native
Platforms	9 (mobile + web + XR)	20+ (game focus)	Android, web, desktop	Web only	Web only	Apple only
Learning curve	Low (Compose/SwiftUI)	High (C#, ECS)	High (C++/JNI)	Medium (JS)	Medium (JS)	Medium (Swift)
AI code gen	MCP server	No MCP	No MCP	No MCP	No MCP	No MCP
Renderer	Filament + RealityKit	Custom (proprietary)	Filament	WebGL/WebGPU	WebGL/WebGPU	Metal

Filament is SceneView's rendering engine on Android and Web. SceneView adds the declarative API, scene graph, AR integration, and cross-platform abstractions on top.

Developer tools

Debug, inspect, and replay your 3D and AR scenes in real time.

bug_report New in v4

AR Debug — Rerun.io

Stream ARCore and ARKit frame data to the Rerun viewer. Inspect camera poses, tracked planes, point clouds, and anchors — then scrub the timeline to replay any moment. Hit Save & Share in the demo to attach a fully-replayable session to a bug report.

Android iOS Python sidecar

Open hosted viewer → Open AR Rerun in the demo app →

videocam v4

AR Record & Replay

Capture an outdoor ARCore session once with rememberARRecorder(). Replay it 1:1 at the desk via ARSceneView(playbackDataset = file). Iterate on AR demos without leaving your chair.

Android 21 unit tests

straighten v4

Image Stabilization (EIS)

One-line opt-in to ARCore Electronic Image Stabilization for shake-free AR previews on supported devices. ARSceneView(imageStabilizationMode = EIS).

Android

cloud_sync

Cloud Anchors

Persistent AR experiences across devices, sessions, and OSes. CloudAnchorNode.host(ttlDays = N) and .resolve(id). Build multi-user AR by passing a string.

Android iOS

analytics