Digital Twins, 3D Modeling & Simulation MCP Servers — CAD, Physics, Game Engines, and Engineering Simulation

Digital twins, 3D modeling, and simulation MCP servers let AI agents control CAD tools, run physics simulations, manipulate game engines, and interact with engineering models. Instead of manually scripting 3D scenes or configuring simulation parameters, AI assistants can directly create models, run simulations, and analyze results through natural language.

This review covers digital twins, 3D modeling, and simulation MCP servers — 3D creation suites, CAD platforms, game engines, physics simulators, and engineering tools. For related servers, see our Robotics & Automation review, Data Science & Machine Learning review, and Scientific Research review.

The headline findings: Blender leads the entire MCP ecosystem — ahujasid/blender-mcp has 26.3k stars, making it one of the most popular MCP servers period. Game engines are well-served, and unevenly so — CoplayDev/unity-mcp has grown to 13.7k stars, well ahead of the other Unity and Unreal MCPs (1.1k–4.0k stars). CAD covers open-source tools with FreeCAD (1.9k stars) and OpenSCAD well-represented. Physics simulation ranges from educational to research-grade with GPU-accelerated quantum and molecular dynamics. Enterprise engineering tools are the gap — no ANSYS, COMSOL, SolidWorks, or cloud digital twin MCPs exist yet.

3D Modeling — Blender

ahujasid/blender-mcp — Most Popular 3D MCP Server

One of the most popular MCP servers in the entire ecosystem — connects Blender to Claude AI through a socket-based server for comprehensive 3D control:

• Object manipulation — create, modify, and delete 3D objects with natural language
• Material control — create and apply materials, adjust properties
• Scene inspection — retrieve detailed scene hierarchy and object information
• Python execution — run arbitrary Python code inside Blender for anything the tools don’t cover
• Poly Haven integration — download free HDRIs, textures, and 3D assets directly
• Hyper3D Rodin — generate AI 3D models from text descriptions
• Hunyuan3D — additional AI 3D model generation pipeline
• Sketchfab integration — search and download 3D models from Sketchfab’s library
• Remote host support — connect to Blender running on a different machine
• Viewport screenshots — capture scene views for AI analysis
• Anonymous telemetry — usage tracking (can be disabled)

The gold standard for AI-assisted 3D modeling. Works with Claude Desktop, Cursor, and any MCP client.

namurokuro/Blender-MCP-Server — Agent-Orchestrable Blender Server

Turns Blender into a fully orchestrable MCP server — 50+ tools exposed over HTTP endpoints, designed for multi-agent pipelines rather than single-user chat. (Previously listed under a poly-mcp GitHub organization; that org no longer exists on GitHub — this is the current live repo matching the same description, now with 10 forks including jlpschell/blender-mcp-server.)

• 50+ tools — comprehensive coverage of Blender operations
• HTTP transport — works over network, not just local stdio
• Thread-safe execution — handles concurrent requests safely
• Auto-dependency installation — sets up its own requirements
• PolyMCP integration — designed for intelligent tool orchestration across multiple servers

Better choice for agentic workflows where multiple AI agents need to coordinate 3D operations.

Other Blender MCPs

• dhakalnirajan/blender-open-mcp — runs with local Ollama models instead of cloud APIs, useful for offline or privacy-sensitive workflows
• VxASI/blender-mcp-vxai (298 stars, MIT) — enhanced Blender MCP with additional features for VXAI platform

CAD — FreeCAD, OpenSCAD, and Cloud CAD

neka-nat/freecad-mcp — Most Popular FreeCAD MCP

FreeCAD control through Claude AI — runs as a FreeCAD addon with an RPC server for conversational 3D design:

• 12 tools — document creation, object creation/editing/deletion, Python execution, screenshots, parts library access, FEM analysis
• FreeCAD addon — installs directly into FreeCAD’s addon system
• Remote connection — connect to FreeCAD running on a different machine
• Parts library — access and insert pre-built components
• 2D to 3D — supports creating objects from 2D drawings

Solid choice for open-source CAD workflows. Design 3D parts through natural conversation.

Other FreeCAD MCPs

• proximile/FreeCAD-MCP — Docker containerization, Vision AI analysis, multi-model support (Claude, GPT-4o, Gemini)
• spkane/freecad-addon-robust-mcp-server — 150+ tools organized by category, the most comprehensive FreeCAD MCP
• bonninr/freecad_mcp — prompt-assisted CAD design, Claude Desktop integration
• contextform/freecad-mcp — open-source FreeCAD automation server
• ATOI-Ming/FreeCAD-MCP — GUI panel + MCP server for running macros and managing documents

jhacksman/OpenSCAD-MCP-Server — Parametric 3D from Text and Images

Text and image to parametric 3D models — goes beyond simple OpenSCAD scripting with AI image generation and multi-view reconstruction:

• AI image generation — create reference images from text descriptions
• Multi-view reconstruction — CUDA-accelerated stereo reconstruction from multiple views
• Parametric export — CSG, AMF, 3MF, and SCAD output formats
• Image approval workflow — review generated images before 3D reconstruction
• Remote processing — offload heavy computation to remote GPU servers

Other OpenSCAD and Cloud CAD MCPs

• quellant/openscad-mcp — render, export, and analyze 3D models using OpenSCAD
• fboldo/openscad-mcp-server — STL and PNG rendering for iterative agent-driven CAD workflows
• jabberjabberjabber/openscad-mcp — LLM-driven OpenSCAD rapid prototyping
• BuildCAD AI (buildcad.ai/mcp) — free cloud-based CAD with multi-view PNG renders (front/right/top/isometric), works with any MCP client
• Svetlana-DAO-LLC/cad-agent — build123d + VTK rendering in Docker, STL/STEP/3MF export, printability analysis
• daobataotie/CAD-MCP — general CAD MCP server

CAD — Fusion 360

Joe-Spencer/fusion-mcp-server — Fusion 360 Design Resources

Exposes Fusion 360’s design hierarchy as MCP resources — rather than just executing commands, this server provides contextual information that helps AI assistants understand your design:

• Design structure — hierarchical view of components, bodies, and sketches
• Parameters — access user-defined parameters for parametric modeling
• Document info — metadata about the currently open file
• ADSK integration — resources and tools from Autodesk’s ecosystem

Other Fusion 360 MCPs

• mycelia1/fusion360-mcp-server — generate and execute scripts directly in Fusion 360
• AuraFriday/Fusion-360-MCP-Server — control Fusion 360 with any AI through MCP
• ArchimedesCrypto/fusion360-mcp-server — Fusion 360 control and automation

Game Engines — Unity

IvanMurzak/Unity-MCP — Editor + Runtime Unity Control

AI-powered bridge for both Unity Editor and compiled games — a comprehensive Unity MCP, with 70+ built-in tools across four categories:

• Project & Assets (21 tools) — manage assets, scripts, and project structure
• Scene & Hierarchy (23 tools) — create, modify, reparent, transform GameObjects
• Scripting & Editor (13 tools) — C# code compilation via Roslyn, editor automation
• Profiling & Diagnostics (12 tools) — performance monitoring, memory analysis, rendering statistics
• Runtime support — works in compiled games, not just the editor — enables real-time AI debugging and player-AI interaction
• Extensible — add custom MCP Tools, Resources, and Prompts
• Multi-deployment — local stdio, remote HTTP, Docker, or standalone binary
• Extensions — Animation, Cinemachine, Navigation, ParticleSystem, ProBuilder, Splines, Terrain, Tilemap, and Timeline modules available

A strong choice for production Unity workflows that need editor and runtime AI integration, though CoplayDev/unity-mcp (below) is now the most-starred Unity MCP by a wide margin.

CoderGamester/mcp-unity — WebSocket Unity Bridge

WebSocket bridge between Unity Editor and AI assistants — a Node.js server implementing MCP that connects to Unity’s editor:

• 30+ tools — scene/GameObject manipulation, component management, material creation, package management, test execution
• MCP resources — query menus, hierarchy, GameObjects, logs, packages, assets
• IDE integration — automatic workspace integration with Unity’s PackedCache folder
• Multi-platform support — Cursor, Windsurf, Claude Code, GitHub Copilot

Strong alternative with emphasis on IDE integration and workspace awareness.

Other Unity MCPs

• CoplayDev/unity-mcp (13,700 stars, MIT) — asset management, scene control, script editing, task automation. By star count this is the most popular Unity MCP server, well ahead of IvanMurzak/Unity-MCP and CoderGamester/mcp-unity above.
• akiojin/unity-mcp-server — deprecated; the README states “development has moved to akiojin/unity-cli”

Game Engines — Unreal Engine

chongdashu/unreal-mcp — Most Popular Unreal Engine MCP

Natural language control of Unreal Engine 5 — a C++ plugin for TCP communication with Unreal Editor subsystems plus a Python MCP server:

• Actor management — create, delete, and transform actors in the scene
• Blueprint development — create custom components with visual scripting
• Blueprint node graph — edit and visualize blueprint execution flow
• Viewport control — manage editor camera and viewport settings
• Multi-client support — Cursor, Claude Desktop, Windsurf

The most mature Unreal Engine MCP with the strongest community adoption. The project’s own README describes it as being in an experimental state: “Breaking changes may occur without notice” and “Production use is not recommended at this time.”

flopperam/unreal-engine-mcp — World Building and Architecture

Specialized for world building — create entire towns, castles, mansions, and mazes with AI-powered commands:

• Blueprint visual scripting — 23+ node types, variables with full property control, custom functions
• Architecture tools — arches, staircases, towers, and large-scale structures with configurable parameters
• Maze generation — recursive backtracking algorithm for procedural maze creation
• Physics & materials — physics simulation controls and material application
• Real-time chat — embedded browser in Unreal Editor for agent interaction

The project’s README now leads with “Flopperam has been acquired by [Aura],” directing users toward a hosted service (tryaura.dev) with an expanded toolset for the most advanced features, while the original open-source local MCP server remains available in this repo under the MIT license with a smaller, foundational toolset.

Other Unreal Engine MCPs

• ayeletstudioindia/unreal-analyzer-mcp — Unreal Engine 5 project analysis
• ChiR24/Unreal_mcp — C++ Automation Bridge plugin with TypeScript server
• kvick-games/UnrealMCP — lightweight agent control for Unreal Engine

Physics Simulation

IBM/chuk-mcp-physics — Comprehensive Physics Calculator

55 specialized physics tools spanning classical mechanics to fluid dynamics — the most complete physics calculation MCP server:

• Basic mechanics (8 tools) — projectile motion, forces, energy, momentum, collisions
• Fluid dynamics (10 tools) — drag, buoyancy, terminal velocity, lift, Magnus force, Bernoulli principle
• Rotational dynamics (5 tools) — torque, moment of inertia, angular momentum
• Oscillations (5 tools) — springs, pendulums, harmonic motion, damping
• Circular motion (5 tools) — centripetal force, orbits, banking angles, escape velocity
• Statics (7 tools) — force balance, torque balance, friction, beam reactions
• Kinematics (7 tools) — motion analysis, trajectory fitting, velocity calculations
• Collisions (2 tools) — elastic and inelastic 3D collisions with energy loss
• Conservation laws (4 tools) — energy, momentum, angular momentum verification
• Unit conversions (2 tools) — 62 unit types across 16 categories
• 515 tests at 98% code coverage — well-tested and reliable

Originally published by chrishayuk; the repository has since been transferred to the IBM GitHub organization (chrishayuk/chuk-mcp-physics now redirects there). Excellent for education, engineering calculations, and physics verification.

andylbrummer/math-mcp — GPU-Accelerated Scientific Simulation

Four specialized GPU-accelerated MCP servers for computational science:

• Math MCP (14 tools) — symbolic algebra, numerical computing
• Quantum MCP (12 tools) — wave mechanics, Schrödinger simulations, double-slit interference
• Molecular MCP (15 tools) — classical molecular dynamics, N-body dynamics, crystal diffraction
• Neural MCP (16 tools) — neural network training and evaluation
• GPU acceleration — automatic CUDA detection, 60-120x speedup over CPU
• Async support — long-running simulations as background tasks

Research-grade simulation capabilities. The quantum and molecular servers enable real scientific computation through AI.

Genesis Physics Engine + MCP Wrapper

Genesis is a next-generation physics engine. The project’s creators report simulating a manipulation scene with a single plane and a Franka robotic arm at up to 43 million FPS — roughly 430,000x real time — though this figure isn’t in the current repo README. It should also be read with a significant caveat: an independent benchmark audit by robotics researcher Stone Tao (creator of the ManiSkill3 simulator) found the 43M FPS number comes from a benchmark where the robot performs one action and then sits idle for over 90% of the test. Correcting the benchmark to use realistic physics substeps, continuous robot motion, and enabled self-collision detection dropped the same Franka-arm scene to roughly 0.29 million FPS — about a 150x reduction — and Tao found Genesis running 3-10x slower than existing GPU simulators like ManiSkill/SAPIEN on realistic manipulation tasks. Treat “43 million FPS” as a best-case, largely-idle-robot number rather than a general performance claim. The underlying engine (renamed since launch from Genesis-Embodied-AI/Genesis to Genesis-Embodied-AI/genesis-world; the old URL redirects) is nonetheless one of the most popular open-source physics projects:

• Multiple solvers — rigid body, MPM, SPH, FEM, PBD, and Stable Fluid
• Material diversity — rigid bodies, liquids, gases, deformable and soft objects
• Robot support — robotic arms, legged robots, drones, soft robots
• Differentiable — supports gradient-based optimization
• Cross-platform — Linux, macOS, Windows with multiple compute backends

The MCP wrapper (dustland/genesis-mcp) is minimal — only 1 tool (run_simulation) plus a world-info resource, exposed via stdio transport. Significant opportunity for a more comprehensive MCP integration given the engine’s capabilities.

Other Physics MCPs

• manasp21/PsiAnimator-MCP — QuTiP quantum physics + Manim animation for educational visualizations
• pathintegral-institute/mcp.science — DFT calculations, materials science data, sandboxed Python execution for research

Engineering Simulation

omni-mcp/isaac-sim-mcp — NVIDIA Isaac Sim Robotics Simulation

Natural language control of NVIDIA Isaac Sim — bridge between conversational AI and precision robotics simulation:

• Scene management — create physics scenes with configurable gravity, timestep, and solver iterations
• Robot placement — position Franka, Jetbot, Carter, G1, and Go1 robots at specified coordinates
• Script execution — run arbitrary Python within Isaac Sim for advanced workflows
• Omniverse commands — execute Omniverse Kit commands directly
• Interactive preview — preview code before execution for safety

Requires NVIDIA Isaac Sim 4.2.0+ and a compatible GPU. Purpose-built for robotics research and digital twin validation.

Orthogonalpub/modelica_simulation_mcp_server — Multi-Physics Modeling

Transforms the Modelica ODE IDE into an AI engineering agent — describe a physical system in natural language, get differential equations, simulation, and visualization:

• Natural language to ODE — automatically generates Modelica models from text descriptions
• Diagram generation — intelligent component placement and connection visualization
• Real-time plotting — live simulation results with 3D visualization
• Cloud execution — token-based authentication with secure cloud compute
• End-to-end automation — from description to simulation results in one prompt

Particularly interesting for multi-domain engineering (mechanical, electrical, thermal, hydraulic systems).

clanker-lover/spicebridge — Circuit Simulation

ngspice circuit simulation through AI — full electronics design workflow:

• 28 tools — template-based design, netlist creation, simulation, measurement, schematic generation
• Simulation types — AC, transient, and DC analysis
• Auto-calculation — component values calculated from templates
• Spec verification — automated measurement against specifications
• Dual transport — stdio for Claude Code, HTTP + Cloudflare tunnel for remote clients

Other Engineering MCPs

• jordy33/iot_mcp_server — IoT device control and monitoring for smart lights, sensors, and connected devices
• ToxMCP/pbpk-mcp — PBPK pharmacokinetic modeling workflows
• globus-labs/science-mcps — Globus compute facilities for scientific simulation at scale

What’s Missing

The biggest gaps in this category:

• No commercial FEA/CFD tools — ANSYS, COMSOL, Abaqus, and NASTRAN have no MCP servers. This is the most impactful gap for professional engineering.
• No SolidWorks MCP — despite SolidWorks’ dominant market share in mechanical engineering, no MCP integration exists.
• No cloud digital twin platforms — Azure Digital Twins, AWS IoT TwinMaker, and similar services have no MCP servers.
• Genesis MCP wrapper is minimal — the physics engine has 29.8k stars but the MCP wrapper only has 1 tool. Significant untapped potential.
• No dedicated CFD server — computational fluid dynamics is entirely absent from the MCP ecosystem.
• No structural analysis — no MCP server handles FEA for stress, strain, or fatigue analysis.
• Limited Autodesk coverage — Fusion 360 has MCPs but AutoCAD, Revit, and Inventor are largely unserved (Autodesk recently announced official MCP servers but they’re in early stages).

What’s New (May 2026)

NVIDIA Omniverse DSX Blueprint reaches general availability (March 16, 2026). NVIDIA announced the Vera Rubin DSX AI Factory reference design and general availability of the Omniverse DSX Blueprint on that date, designed for large-scale AI factory design and digital twin simulation — power, cooling, software, and operations modeled together. Partner integrations cover the full AI factory stack. For the MCP category, this is relevant context: enterprise-grade digital twin infrastructure is maturing rapidly, but MCP integration with Omniverse is still a gap.

Forward Networks: 54 tools for network digital twins (2026). Forward Networks published an official MCP server exposing 54 tools (plus 6 workflow prompts) for network topology analysis, path tracing, device management, and configuration auditing against its network digital twin platform. This represents a new class of “network digital twin” — modeling network topology and behavior rather than physical objects — that sits between traditional physical simulation and IT infrastructure monitoring. NetworkOps_Platform offers a similar community-built approach with 178 tools supporting Cisco, Juniper, Nokia, Arista, and Linux with RAG-powered chat and real-time topology visualization.

Cloud digital twin gap beginning to close. The “What’s Missing” section below flags Azure Digital Twins and AWS IoT TwinMaker as absent from the MCP ecosystem. In January 2026, nCircle Tech documented how an MCP server layer can unify Autodesk applications (BIM 360, ACC, Forge) and partner solutions into a connected, cloud-agnostic infrastructure, moving digital twins from static snapshots toward continuously updated models. No official Azure or AWS digital twin MCP servers have shipped, but the architectural blueprints are now widely published.

Summary

Overall rating: 4/5 — Blender and game engine MCPs are world-class. CAD and physics simulation are solid. Enterprise engineering simulation is the clear gap — when ANSYS, COMSOL, and SolidWorks get MCP integrations, this category will be transformative.

This review is AI-generated based on research of publicly available GitHub repositories and documentation. ChatForest does not test or run MCP servers hands-on. Star counts are approximate and may have changed since publication. Last updated: August 2026.

Category: Design & Creative MCP Servers

This review was last edited on 2026-08-26 using Claude Sonnet 5 (Anthropic).