3D Agent MCP
MCP server for generating print-ready 3D models from text descriptions via a multi-agent pipeline, enabling AI assistants to create and optimize STL files.
README
<div align="center">
🖨️ 3D Agent MCP
Text → 2D Preview → 3D Model → Print-Ready STL
AI-powered multi-agent pipeline for generating 3D printable models from text descriptions, with MCP server for seamless AI assistant integration.
🇬🇧 English | 🇷🇺 Русский
<img src="docs/images/v2_text_to_3d_pipeline.png" alt="Pipeline" width="90%"/>
</div>
Demo
<video src="https://github.com/teslaproduuction/3d-agent-mcp/raw/main/docs/demo.mp4" controls width="100%"></video>
Generated Examples
<table> <tr> <td><img src="docs/PR/data/images/preview_preview_022218a9.png" width="200"/></td> <td><img src="docs/PR/data/images/preview_preview_02a7f736.png" width="200"/></td> <td><img src="docs/PR/data/images/preview_preview_037a6c8e.png" width="200"/></td> <td><img src="docs/PR/data/images/preview_preview_06079652.png" width="200"/></td> </tr> <tr> <td><img src="docs/PR/data/images/preview_preview_066fe8d4.png" width="200"/></td> <td><img src="docs/PR/data/images/preview_preview_08134311.png" width="200"/></td> <td><img src="docs/PR/data/images/mv_back-right_0a51fab9.png" width="200"/></td> <td><img src="docs/PR/data/images/mv_left.png" width="200"/></td> </tr> </table>
2D previews generated before 3D conversion — faster iteration, less API cost
Multi-View Generation
<table> <tr> <td><img src="docs/PR/data/images/preview_zero123_right_bf464842.png" width="150"/></td> <td><img src="docs/PR/data/images/preview_zero123_right_bf93635c.png" width="150"/></td> <td><img src="docs/PR/data/images/preview_zero123_right_c023b3c2.png" width="150"/></td> <td><img src="docs/PR/data/images/preview_zero123_right_c46b046f.png" width="150"/></td> <td><img src="docs/PR/data/images/preview_zero123_right_d243190d.png" width="150"/></td> </tr> </table>
Multiple camera angles → higher-quality 3D geometry via Hunyuan3D-2mv
Architecture
System Context (C4 Level 1)
<img src="docs/images/c4_l1_context.png" alt="C4 Context" width="80%"/>
Containers (C4 Level 2)
<img src="docs/images/c4_l2_containers.png" alt="C4 Containers" width="80%"/>
Agent Pipeline
User Prompt
│
▼
┌─────────────────────┐
│ Planner Agent │ ← Decomposes prompt into objects
└─────────┬───────────┘
│
▼
┌─────────────────────┐
│ Image Gen Agent │ ← DALL-E 3 / FLUX / Qwen (2D preview)
└─────────┬───────────┘
│
[User confirms preview]
│
▼
┌─────────────────────┐
│ Generation Agent │ ← Tripo3D API / Hunyuan3D (local)
└─────────┬───────────┘
│
▼
┌─────────────────────────────────────┐
│ Intelligent PostProcessing Agent │
│ ├── Overhang analysis (24 angles) │
│ ├── Support strategy decision │
│ └── Optimal orientation on bed │
└─────────┬───────────────────────────┘
│
▼
Print-Ready STL
Technical Stack
<img src="docs/images/v2_tech_stack.png" alt="Tech Stack" width="80%"/>
Features
| Feature | Description |
|---|---|
| Text-to-3D | Generate 3D model from any text description |
| 2D Preview gate | Create image preview before expensive 3D API call |
| Intelligent post-processing | AI agent analyzes geometry, decides supports and orientation |
| Multi-view generation | Multiple camera angles → better 3D quality |
| Multi-object scenes | Plan and generate complex scenes with multiple objects |
| MCP integration | Use from Claude Desktop, Cursor, and any MCP client |
| Local models | Hunyuan3D-2, TripoSR, FLUX — no API costs, runs on-premise |
| Docker stack | Full local stack with GPU support |
Intelligent Post-Processing Output
desk_organizer analysis:
✅ Printable without supports in recommended orientation.
Complexity: EASY
AI Analysis:
- Geometry complexity: MEDIUM
- Max overhang angle: 38.5°
- Bed contact area: 1 250 mm²
- No internal cavities detected
- Recommended: rotate 180° around X axis
Quick Start
Option 0 — Docker image (fastest)
docker pull ghcr.io/teslaproduuction/3d-agent-mcp:latest
cp .env.example .env
# Fill in API keys
docker-compose up -d
# → http://localhost:7860
Option 1 — UV (recommended, 10–100× faster than pip)
# Install UV
winget install --id=astral-sh.uv -e # Windows
curl -LsSf https://astral.sh/uv/install.sh | sh # Linux/macOS
# Clone and setup
git clone https://github.com/teslaproduuction/3d-agent-mcp.git
cd 3d-agent-mcp
uv venv --python 3.10
uv sync --all-extras
# Configure
cp .env.example .env
# Edit .env with your API keys
# Run
uv run python ui/gradio_app.py
# → http://localhost:7860
Option 2 — Docker (full stack with local models)
cp .env.example .env
# Edit .env
docker-compose up -d --build
# → http://localhost
Option 3 — pip
python -m venv .venv
source .venv/bin/activate # Linux/macOS
.venv\Scripts\activate # Windows
pip install -r requirements.txt
cp .env.example .env
python ui/gradio_app.py # → http://localhost:7860
MCP Integration
Works with Claude Desktop, Cursor, Windsurf, and any MCP-compatible client.
Claude Desktop config (claude_desktop_config.json)
{
"mcpServers": {
"3d-agent": {
"command": "python",
"args": ["/path/to/3d-agent-mcp/mcp_server/server.py"],
"env": {
"TRIPO_API_KEY": "your_key",
"OPENAI_API_KEY": "your_key"
}
}
}
}
Usage in Claude
User: Generate a phone stand for 3D printing
Claude: [calls generate_3d_model tool]
✅ Model generated and optimized for printing!
- File: outputs/models/phone_stand_optimized.stl
- Supports: none required
- Orientation: base-down
- Print time: ~2h 15min
Available MCP tools: generate_3d_model · generate_2d_preview · analyze_printability · plan_scene
→ See mcp_server/README.md for full API docs.
API Keys
| Key | Purpose | Required |
|---|---|---|
OPENAI_API_KEY |
DALL-E 3 image gen + GPT for agents | For cloud mode |
TRIPO_API_KEY |
3D generation (Tripo3D cloud) | For cloud mode |
ANTHROPIC_API_KEY |
Claude models as agent LLM | Optional |
REPLICATE_API_TOKEN |
SDXL / Flux image generation | Optional |
No cloud keys needed for local mode — run Hunyuan3D + FLUX via Docker stack.
Configuration
config.yaml controls all behavior:
default_settings:
# Image generation
image_generation:
provider: "local" # local | dalle3 | sdxl | flux
# 3D generation
generation:
api_provider: "local" # local | tripo | meshy
face_limit: 10000
# Post-processing
postprocessing:
mode: "intelligent" # AI decides automatically
auto_orient: true
max_overhang_angle: 45.0
# Printer profile
printer:
build_volume: [220, 220, 250] # mm — Ender 3 / Bambu A1
nozzle_diameter: 0.4
material: "PLA"
# LLM backend
llm:
default_provider: "ollama" # ollama | openai | anthropic
local:
ollama_models: ["qwen2.5:32b", "qwen2.5:7b"]
Project Structure
3d-agent-mcp/
├── agents/ # AI agents
│ ├── coordinator.py # Pipeline orchestrator
│ ├── planner_agent.py # Scene decomposition
│ ├── image_generation_agent.py # 2D preview
│ ├── generation_agent.py # 3D API calls
│ └── intelligent_postprocessing_agent.py
│
├── api_clients/ # API wrappers
│ ├── llm_client.py # OpenAI / Anthropic / Ollama
│ ├── image_api_client.py # DALL-E / SDXL / FLUX
│ └── tripo_client.py # Tripo3D
│
├── mcp_server/ # MCP server
│ ├── server.py # Tool definitions
│ └── README.md # MCP API docs
│
├── ui/ # Gradio web UI
│ ├── gradio_app.py # Main app
│ └── tabs/, handlers/, components/
│
├── postprocessing/ # Geometry analysis
├── docker/ # Local model containers
│ ├── hunyuan3d/ # Hunyuan3D-2 (local 3D)
│ ├── flux/ # FLUX.1 (local image gen)
│ ├── comfyui/ # ComfyUI
│ └── nginx/ # Reverse proxy
│
├── tests/
├── config.yaml # Main config
├── .env.example # API key template
├── docker-compose.yml # Full Docker stack
└── pyproject.toml
Diagrams
| Diagram | File |
|---|---|
| Component | docs/PR/diagrams/01_component.png |
| Sequence | docs/PR/diagrams/02_sequence.png |
| Activity | docs/PR/diagrams/03_activity_gci.png |
| Deployment | docs/PR/diagrams/04_deployment.png |
| Classes | docs/PR/diagrams/05_classes.png |
Development
# Run tests
pytest tests/
# Format
black .
# Lint
flake8 .
# Type check
mypy .
Roadmap
- [ ] Meshy API integration
- [ ] PySLM — physics-based support generation
- [ ] G-code preview before printing
- [ ] Printer preset library (Ender 3, Bambu, Prusa)
- [ ] Export to OBJ, FBX, GLTF
- [ ] REST API mode (no Gradio dependency)
Contributing
- Fork the repo
- Create a feature branch:
git checkout -b feature/my-feature - Commit changes:
git commit -m "feat: add my feature" - Push:
git push origin feature/my-feature - Open a Pull Request
License
MIT © 2026 — see LICENSE
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。