SolidworksMCP
Click on "Install Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@SolidworksMCPCreate a sketch on the top plane and extrude it 25mm"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
SolidworksMCP
SolidworksMCP is a production-oriented Model Context Protocol server for controlling a live SOLIDWORKS session through the Windows COM API.
Current release: 6.5.31. The release was acceptance-tested with SOLIDWORKS 2026 SP2.1 and retains compatibility guards for SOLIDWORKS 2025.
Highlights
Native sketch, feature, body, reference-geometry, view, probe, and export operations.
Transactional CAD plans with checkpoints, invariants, idempotency, budgets, and rollback.
Parametric sketches with named entities, constraints, dimensions, equations, persistent IDs, and degree-of-freedom analysis.
Multibody tools with explicit merge scope, clearance, interference, bounding box, and volume verification.
Deterministic raster-to-sketch pipelines for silhouettes and line art, including GPU-backed segmentation, topology preservation, primitive fitting, and reverse-raster quality gates.
Native sketch/body export and calibrated geometric comparison instead of screenshot-only acceptance.
Verified active-sketch Normal To and Fit to Screen. The server reads the resulting camera matrix and pixel occupancy instead of trusting a UI command.
Structured diagnostics, modal-dialog preflight, a narrowly scoped Modify watchdog, and finite recovery policies.
Related MCP server: SolidworksMCP-python
Requirements
Windows 10 or Windows 11.
SOLIDWORKS 2025 or 2026, installed and licensed.
Python 3.10 or newer; Python 3.11 is the acceptance-tested runtime.
Git LFS for the packaged line-art checkpoints.
An NVIDIA CUDA environment is strongly recommended for the deep vectorization modes.
get_capabilitiesreports the actual backend state.
The server drives the interactive SOLIDWORKS process. It is not a headless CAD kernel, and an unknown modal dialog must be inspected by a human.
Clone and install
Install Git LFS before cloning, or pull the model objects afterwards:
git lfs install
git clone https://github.com/czuryk/SolidworksMCP.git
Set-Location SolidworksMCP
git lfs pullCreate an isolated Python environment and install dependencies:
py -3.11 -m venv .venv
.\.venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
python -m pip install -r requirements.txtFor CUDA vectorization, install the PyTorch build appropriate for the machine's
CUDA driver before or together with requirements.txt.
MCP configuration
Point the MCP client at the repository entry point using the virtual environment's Python executable:
{
"mcpServers": {
"SolidworksMCP": {
"command": "C:\\path\\to\\SolidworksMCP\\.venv\\Scripts\\python.exe",
"args": [
"C:\\path\\to\\SolidworksMCP\\solidworks_mcp_server.py"
]
}
}
}Start SOLIDWORKS, restart or reconnect the MCP client, then call
get_capabilities and get_environment_status before a substantial modeling
operation.
Stable compatibility identifiers
SolidworksMCP is the public project and MCP handshake name. The Python package
remains solidworks_mcp, the configured tool namespace remains solidworks,
and existing artifact schemas retain their solidworks-mcp/... identifiers.
Those internal names are compatibility contracts and are intentionally not
renamed.
Operation families
The server exposes both small native operations and higher-level verified workflows. Important families include:
documents and environment: create/open/save/close, capabilities, session metrics, modal-state inspection, recovery;
sketches: standard planes, planar faces, exact model-coordinate contours, primitive drawing, parametric sketches, dimensions, constraints, equations;
features: verified extrude/cut, revolve, sweep, fillet, chamfer, shell, mirrors, patterns, reference axes and planes;
bodies: visibility, naming, appearance, volume, clearance, interference, multibody insert workflows;
verification: ray/section probes, sketch topology, native geometry export, sketch-to-sketch and calibrated image comparisons, body silhouettes;
vectorization: selectable silhouette/region/line-art modes, deep segmentation, controllable approximation, CAD commit, and reverse-raster verification;
orchestration: transactions, declarative CAD plans, semantic primitives, model-graph synchronization, atomic export bundles.
Use MCP tool discovery for the authoritative live schema. Tool descriptions and
get_capabilities are versioned with the server.
Safety rules
Supply an absolute
save_pathbefore the first mutation of an unsaved document, or explicitly opt into disposable work withallow_unsaved_document=true.Treat
successas execution evidence. Verify topology, body count, bounding boxes, volume, tolerances, and a readable independent artifact.Use
scope_bodiesandexpected_merge_bodiesfor multibody cuts and bosses.Do not auto-confirm unknown dialogs. Only the recognized, operation-scoped dimension Modify dialog may be handled by the watchdog.
Native MCP tool coordinates use the declared user unit, normally millimetres. Raw SOLIDWORKS COM coordinates use metres.
Do not invoke a blind
Ctrl+8after sketch operations. Version 6.5.31 already applies and verifies the stable side/up orientation and working-geometry fit.
Vectorization models
The deterministic line-art checkpoints live under
solidworks_mcp/models/lineart/ and are stored with Git LFS:
dexined_biped_v2.ptteed_biped_5.pt
Their source and license notices are documented in
solidworks_mcp/models/lineart/LICENSES.md. A clone containing small LFS pointer
files instead of the binary objects is incomplete; run git lfs pull before
starting the server.
SOLIDWORKS Autotrace/Picture to Sketch is an interactive PropertyManager feature and has no supported public API entry point. This server therefore uses its own deterministic segmentation and line-art pipelines.
Configuration
Default runtime settings are stored in solidworks_mcp/config.json. Keep
machine-specific or sensitive overrides in config.local.json; that file is
ignored by Git.
Notable defaults include:
automatic SOLIDWORKS executable and template discovery;
millimetres as the user unit;
operation-scoped dimension dialog guard;
bounded transaction/runtime limits;
automatic checkpoint directory management;
INFO-level file logging.
Relative log_file values are written under
%LOCALAPPDATA%\SolidworksMCP\, not into the source checkout. An absolute path
in config.json remains supported.
Tests
The repository contains the deterministic unit and regression suite:
python -m unittest discover -s tests -p "test_*.py"Release 6.5.31 passes 155 tests. These tests do not replace live acceptance for COM behavior: qualifying another SOLIDWORKS major version should also cover standard planes, planar faces, existing-sketch activation, view recovery, features, transactions, vectorization, and export.
Project layout
SolidworksMCP/
|-- solidworks_mcp_server.py MCP stdio entry point
|-- requirements.txt
|-- LICENSE
|-- solidworks_mcp/
| |-- server.py MCP schemas and dispatch
| |-- tool_registry.py high-level tool metadata
| |-- vector_worker.py isolated scientific worker
| |-- config.py / config.json
| |-- constants.py
| |-- automation/ SOLIDWORKS COM implementation
| |-- models/lineart/ Git-LFS checkpoints and licenses
| `-- utils/
`-- tests/
`-- test_v6.py unit and regression suiteRuntime logs, caches, local client settings, checkpoints, generated reports, acceptance artifacts, and manual probes are intentionally excluded from the repository.
License
The server is distributed under the MIT License; see LICENSE. Packaged model
notices are retained separately in solidworks_mcp/models/lineart/LICENSES.md.
Original project: alisamsam/Solidworks-MCP.
This server cannot be installed
Maintenance
Resources
Unclaimed servers have limited discoverability.
Looking for Admin?
If you are the server author, to access and configure the admin panel.
Related MCP Servers
- Alicense-qualityBmaintenanceMCP bridge server for SolidWorks that enables AI assistants to control SolidWorks programmatically via COM automation.Last updatedMIT
- AlicenseCqualityCmaintenancePython MCP server for SolidWorks automation with 109 tools covering the full CAD lifecycle. Enables AI-assisted design workflows through COM automation on Windows.Last updated10044MIT
- Alicense-qualityCmaintenanceEnables AI (e.g., Claude) to control SolidWorks via natural language, automating part creation, sketching, and feature operations through the Model Context Protocol.Last updated1MIT
- Alicense-qualityCmaintenanceEnables connecting MCP clients to local SolidWorks for automated 3D part creation, template generation, file export, and basic review via natural language.Last updated1MIT
Related MCP Connectors
MCP server for generating rough-draft project plans from natural-language prompts.
MCP server for Klever blockchain smart contract development.
MCP server providing access to the Scorecard API to evaluate and optimize LLM systems.
Latest Blog Posts
- Who's Calling? MCP Hosts Are an Identity Blind Spot (And the Spec Knows It)By Om-Shree-0709 on .mcpAgent IdentityOAuth 2.1
- Your AI Chatbot Just Exposed Your CEO's Salary to an InternBy Om-Shree-0709 on .Agent IdentityMCP SecurityOAuth Delegation
- Why MCP Servers Need Execution Sandboxing (And Why Your Current Stack Isn't Enough)By Om-Shree-0709 on .Agentic AiPrompt InjectionWebAssembly
MCP directory API
We provide all the information about MCP servers via our MCP API.
curl -X GET 'https://glama.ai/api/mcp/v1/servers/czuryk/SolidworksMCP'
If you have feedback or need assistance with the MCP directory API, please join our Discord server