mcp-server-solidworks

What is mcp-server-solidworks?

An MCP server that lets AI models work with SolidWorks at the CAD feature level. It converts user intent into a CAD-neutral intermediate representation, then a deterministic compiler lowers it into concrete SolidWorks operations. Designed for any MCP-capable AI client (Claude, OpenClaw, OpenAI agents, local LLMs).

How to use mcp-server-solidworks?

Requires Windows, SolidWorks 2026, .NET Framework 4.8, and Python 3.x. Build the C# execution layer with MSBuild, run it as a headless server on http://localhost:5000, then run the Python adapter with FastMCP. Register the adapter with an MCP-capable AI client by adding an entry to its mcpServers config (e.g., Claude Desktop’s claude_desktop_config.json). Restart the client after any config change.

Key features of mcp-server-solidworks

• Raises abstraction from thousands of API methods to feature-level intent.
• CAD-neutral Feature Graph IR as intermediate representation.
• Deterministic compiler lowers IR into ordered SolidWorks operations.
• Open MCP standard; not tied to any single AI client.
• Roughly three dozen low-level tools for modeling, sketching, drawing, and export.
• Contract test keeps adapter and execution layer in exact sync.

Use cases of mcp-server-solidworks

• Automate repetitive CAD part creation (sketches, extrusures, patterns, sheet metal).
• Generate dimensioned multi-view drawings from existing models.
• Reverse-engineer a part by reading a drawing’s 2D geometry.
• Batch export models to STEP, IGES, STL, PDF, DWG, DXF.
• Run CAD workflows with any MCP-capable AI agent.

FAQ from mcp-server-solidworks

What are the runtime and dependency requirements?

Windows, SolidWorks 2026, .NET Framework 4.8, and Python 3.x. The adapter requires the FastMCP package. The execution layer communicates over REST at http://localhost:5000.

How does it avoid the tool explosion problem common in AI+CAD integrations?

It raises the abstraction level: the AI reasons in terms of CAD features (e.g., “put a hole in the top face”), not individual API methods. A deterministic compiler converts that intent into the concrete SolidWorks operations, reducing the number of model calls per request.

Is the Feature Graph IR and compiler ready for production use?

No. The IR schema is drafted and an experimental path exists behind a flag (SOLIDPILOT_ENABLE_IR=true), but the full compiler (especially the reference resolver) is not yet implemented. The current working path uses low-level MCP tools directly.

What transports and authentication does it use?

The adapter communicates with the execution layer via plain REST. MCP sits at the top as the AI-client-facing boundary. No authentication mechanism is mentioned; the server is assumed to run locally.

What is the project status?

Early alpha. Part-modeling tools are most mature. Drawing tools are a maturing capability. The strategic IR+compiler path is not yet built. Section views are experimental and not reliable under automation.