Skip to content

SolidWorks MCP Server

Tried it? Tell us what worked and what didn't in a feedback issue. Reports from real setups decide what gets fixed next.

Python 3.13+ MCP Compatible License: MIT Windows SolidWorks Docs Tests Coverage

AI-Assisted SolidWorks Automation with MCP

🚀 152 Tools | 🤖 Agent-Ready Workflows | ⚡ Automatic COM/VBA Routing | 🧪 Schema-Validated Prompt Runs

Overview

An attempt to practical MCP server for SolidWorks focused on faster iteration with AI: describe intent, generate a plan, execute tools, inspect results, and iterate. This is a differentiating tool from some other really great projects since I still want to use Solidworks, and don't want to change to another 3D modeling engine. This leads to the following project that can work with Solidworks to update/design/create parts in a somewhat interactive manner, and helps add tooling etc. to LLMs to help making designing faster and easier, with the intent of removing the boring and repetitive parts of the 3D modeling process.

It supports both experienced CAD users and newer makers who want a tutor-like assistant for 3D-printable design workflows.

Motivation

This project came from a pretty simple problem: I did not have enough time to keep iterating on my 3D printing hobby projects, especially failed prints/parts I wanted to fix quickly. I also had a couple of SolidWorks courses behind me, but still wanted faster ways to go from idea to printable part with less hassle and iterate faster.

In 2024, I started a VBA script to generate a D-100 die (100-sided dice) for a friend in SolidWorks and tried using SOTA models like Claude Sonnet and ChatGPT for help with the SolidWorks VBA API. The experience was terrible. The models were often confidently wrong, struggled with basic API/version details, and seemed to miss core CAD sequencing (sketch -> feature -> part -> assembly).

Even with around 1,000 hours in SolidWorks, I still have practical workflow questions: should I import a mesh and convert it, or model from scratch, and what approach will stay printable without taking forever? Out-of-the-box prompting did not connect any of those dots.

So this repo is my attempt to close that gap, or at least research what is feasible. It is a hobby/research project, and if these findings help someone else build faster, that is a win. Feel free to open an issue with questions or feedback.

Why This Project Is Useful Now

1) AI-Assisted CAD Iteration That (Tries) to Stays Grounded

The project emphasizes an inspect -> classify -> delegate loop instead of blind generation. In practice this means:

  • read model state first (get_model_info, list_features, get_mass_properties)
  • classify feature family before planning (classify_feature_tree)
  • route to direct MCP or VBA-backed execution based on evidence

2) Broad Tool Surface with Practical Coverage

  • 152 tools across modeling, sketching, drawing, engineering data, analysis, export, automation, templates, macros, and docs discovery
  • one server API surface for both simple operations and advanced workflows

3) Reliable Execution Path for Complex Operations

  • direct COM calls for straightforward operations
  • automatic VBA generation fallback for higher-complexity calls
  • circuit-breaker and connection-pool patterns for stability under repeated use

4) Agentic Layer with Real Validation

  • Pydantic and PydanticAI schema validation for structured outputs
  • recoverable-failure handling and retry flow for prompt-driven runs
  • local SQLite logging and timeline inspection for reproducibility and debugging

5) Lower Barrier for New CAD/3D Printing Users

  • prompt-driven tutorials and worked examples designed for "describe -> review -> build"
  • specialist agents for manufacturability, documentation workflows, and research validation
  • practical orientation for people who know 3D printing goals but are still learning CAD execution

Security and Deployment Controls

Security is included and configurable, but it is not the primary value proposition of this project.

The server still provides multiple security/deployment modes for local development, controlled internal use, and constrained environments when needed.

Current Product Positioning

  • Best fit: AI-assisted SolidWorks automation and workflow acceleration
  • Also good for: reusable tutorials, team playbooks, and prompt-to-tool validation
  • Not positioned as: Standalone Solidworks Agent etc.

Not Yet Supported

  • Mock adapter output is simulated and should not be treated as engineering truth.
  • Live 3D viewport streaming in a UI.
  • Checkpoint-level interference validation.
  • Simulation and fluid workflows.
  • Topology optimization (for example via SimulationXpress).

What You Can Do Quickly

  • build and modify SolidWorks parts from prompt-driven plans
  • audit existing sample parts before reconstruction
  • compare outputs with feature tree and mass-property checks
  • run schema-validated agent prompts and store run history for review

Quick Start

Choose the path that matches your setup:

Windows only

Use this when SolidWorks and the MCP server run on the same Windows machine. This is the verified setup path.

git clone https://github.com/andrewbartels1/SolidworksMCP-python.git
cd SolidworksMCP-python
.\install.ps1

install.ps1 creates .venv, installs the server, then runs solidworks-mcp setup, which finds your SolidWorks and MCP clients and connects the ones you pick. See the Installation guide.

Open SolidWorks, restart your client, and ask "What model is open in SolidWorks?". Healthy server logs include:

  • SolidWorks COM interface is available
  • Registered 156 SolidWorks tools (the count grows as tools are added)
  • Connected to SolidWorks

Linux / WSL only

Use this for mock-mode development, tests, and documentation work.

git clone https://github.com/andrewbartels1/SolidworksMCP-python.git
cd SolidworksMCP-python
make install
make test
make docs

Linux / WSL client + Windows host

Use this when SolidWorks runs on Windows and your client or development workflow runs on Linux/WSL.

.\.venv\Scripts\python.exe -m solidworks_mcp.server --mode remote --host 0.0.0.0 --port 8000

make install
make test

Then connect your client to http://<windows-host-ip>:8000.

Tool Categories

Category Tools Description
Modeling 11 Part creation, features, assemblies
Sketching 19 Complete sketching toolkit with constraints
Drawing 12 Drawing creation and management
Drawing Analysis 8 Quality analysis and compliance checking
Analysis 5 Mass properties, simulation, validation
Export 7 Multi-format export and conversion
Automation 8 Batch processing and workflows
File Management 14 File operations and organization
VBA Generation 10 Dynamic VBA code for complex operations
Template Management 6 Template creation and standardization
Macro Recording 7 Macro recording, optimization, and libraries
Docs Discovery 2 API and COM documentation discovery and lookup

Architecture Overview

The SolidWorks MCP Server uses an intelligent adapter architecture that automatically routes operations between direct COM API calls and VBA macro generation based on complexity analysis:

flowchart TB
    Client["MCP Client"] --> Server["FastMCP Server"]
    Server --> Router["Intelligent Router"]
    Router --> Analyzer["Complexity Analyzer"]

    Analyzer -->|Simple Operations| COM["Direct COM API"]
    Analyzer -->|Complex Operations| VBA["VBA Generation"]

    COM --> SW["SolidWorks Application"]
    VBA --> SW

    Router --> CB["Circuit Breaker"]
    Router --> Pool["Connection Pool"]

Getting Started

Ready to automate your SolidWorks workflows? Check out our comprehensive guides:

Ready to get started? → Installation Guide