Simul: The MCP Bridge Connecting AI Agents to the 3D Simulation World

Introducing the Simul project, an MCP protocol-based bridge for 3D simulation and OpenUSD workflow orchestration, enabling AI Agents to connect design tools, physics engines, and spatial environments, thus ushering in a new era of AI-driven 3D content creation.

Traditional 3D content creation and simulation workflows rely heavily on manual operations:

• Tool fragmentation: Designers need to switch between multiple software like Blender, Maya, Houdini, Unreal Engine, etc.
• Steep learning curve: Mastering these tools requires long-term professional training
• Repetitive work: A lot of time is spent on mechanical tasks such as parameter adjustment, format conversion, and scene setup

The introduction of AI Agents is expected to completely change this situation, but only if the interoperability issue between Agents and the 3D tool ecosystem is resolved.

Simul is built based on the MCP (Model Context Protocol) protocol, an open standard launched by Anthropic aimed at standardizing interactions between AI models and external tools/data sources. The advantages of choosing MCP are:

• Ecosystem compatibility: Natively compatible with mainstream AI assistants like Claude and Cursor
• Standardized interface: Unified tool discovery and calling mechanism
• Security and controllability: Built-in permission management and sandbox mechanism
• Community-driven: Open protocol specifications to avoid vendor lock-in

One of Simul's core technical pillars is OpenUSD (Universal Scene Description). Developed by Pixar, OpenUSD is an industrial-grade 3D scene description standard widely adopted by giants like Apple, NVIDIA, and Adobe:

Layered Scene Graph: OpenUSD uses a layered, composable scene graph structure that supports:

• Non-destructive editing (each modification is overlaid as a new layer)
• Multi-user collaboration (different layers can be edited by different users/tools)
• Variant management (multiple configurations of the same asset)

Cross-tool interoperability: As a universal exchange format, USD breaks down barriers between tools, allowing Simul to uniformly coordinate workflows across multiple software.

Simul encapsulates 3D tools, physics engines, and spatial environments into standard MCP tools, which Agents can call via a unified interface:

Design Tool Connectors:

• Blender: Open-source 3D creation suite supporting modeling, animation, and rendering
• Maya: Industry-standard animation and modeling software
• Houdini: The king of procedural content creation
• 3ds Max: Architectural visualization and game asset production

Physics Engine Integration:

• NVIDIA PhysX: Real-time physics simulation
• Bullet: Open-source physics engine commonly used in robot simulation
• MuJoCo: Standard for reinforcement learning research

Spatial Environment Interfaces:

• Apple Vision Pro spatial computing
• OpenXR cross-platform VR/AR
• NVIDIA Omniverse collaboration platform

Simul is not just a simple tool connector but an intelligent workflow orchestrator:

Intent Understanding Layer: Receives natural language instructions from Agents and parses them into structured task descriptions. For example:

• "Create a scene with 5 spheres and make them fall from a height"
• "Export this character model to FBX format and upload it to the cloud"
• "Optimize the geometric structure of this scene to reduce polygon count while maintaining visual quality"

Task Decomposition Layer: Breaks down complex instructions into executable subtask sequences, determining dependencies and execution order.

Resource Scheduling Layer: Selects appropriate tool instances based on task requirements and manages computing resource allocation.

Execution Monitoring Layer: Tracks task execution status, handles exceptions, and feeds back progress and results to Agents.

Game and film production require a large number of 3D assets. Simul enables AI Agents to:

• Automatically generate buildings, props, and characters according to design specifications
• Create parameterized variant asset libraries
• Intelligently layout scene elements, following composition and gameplay rules

In industrial design and robotics fields:

• Automatically build virtual environments for product testing
• Perform large-scale parameter scanning and optimization
• Generate simulation reports and visualization results