Our CAD tools, inside your assistant
Three conversions you can hand to Claude, ChatGPT or Gemini and then simply ask for: a mesh into a solid STEP file, a 2D PDF drawing into a 3D model, and photographs of a real part into a parametric one. One connection covers all three.
What your assistant gains
- stl_to_step
- A mesh — STL, OBJ, PLY, 3MF, GLB — rebuilt as a solid with real flats, cylinders and holes instead of thousands of facets. No AI involved; usually under a minute.
- drawing_to_cad
- A 2D PDF drawing read for its views, dimensions and notes, built as a 3D model, then checked back against the drawing. The job carries that report, so you can see where it agreed and where it did not.
- photo_to_cad
- Photographs of a part in your hand. It works out the shape, asks for the handful of dimensions a photograph cannot give, and builds a parametric model from your answers. This is the one assistants are unexpectedly good at, because asking the questions is the hard part.
Connecting it
The tools speak Model Context Protocol, which all three assistants support, so there is one address to paste:
https://digiforgemfg.com/mcp
You will need a key. Email sales@digiforgemfg.com and say roughly what you are building; keys are free and carry a daily job allowance.
Send it as Authorization: Bearer dfk_... where your client lets you set a header, or as
?key=dfk_... on the address where it does not. The header is better — a key in a URL tends
to end up in logs — so use the query form only where there is no other way.
Claude
claude.ai or the desktop app: Customize → Connectors → Add → Add custom connector. Name it, give it the address above, and press Continue. On the second screen, two things matter:
- under Authentication, choose No sign-in — this server uses an API key, not OAuth, and the other choices will send you round an OAuth loop that goes nowhere;
- under Request headers, add
Authorizationwith the valueBearer dfk_your_key.
Then press Add. (Putting ?key=dfk_your_key on the address works too, if you would rather
not use a header.)
Claude Code takes the same key as a header:
claude mcp add --transport http digiforge https://digiforgemfg.com/mcp \
--header "Authorization: Bearer dfk_your_key"
ChatGPT
Connectors: Settings → Connectors → add a custom MCP connector with the address above,
authenticating with your key as a bearer token.
A custom GPT instead: in the GPT editor, Actions → Import from URL, and paste
https://digiforgemfg.com/openapi.json. Set Authentication to API Key, type Bearer.
Gemini
Gemini CLI: add this to ~/.gemini/settings.json:
{
"mcpServers": {
"digiforge": {
"httpUrl": "https://digiforgemfg.com/mcp",
"headers": { "Authorization": "Bearer dfk_your_key" }
}
}
}
The Gemini API: the same OpenAPI document, https://digiforgemfg.com/openapi.json, describes
every call as a function declaration.
Anything else
There is a plain REST surface under /api/v1/ for clients that do not speak MCP:
curl -X POST https://digiforgemfg.com/api/v1/stl-to-step \
-H "Authorization: Bearer dfk_your_key" \
-F file=@bracket.stl
curl https://digiforgemfg.com/api/v1/jobs/mesh:JOB-ID \
-H "Authorization: Bearer dfk_your_key"
How files get here
A CAD file is almost always on somebody's own computer, and a tool call carries arguments, not
megabytes. So the assistant calls open_upload, gets a link, and gives it to the person;
they drop the file on it and the conversion goes ahead. Files already at a public address can be passed
as a URL instead. Links last two hours and cannot read anything back.