Rupa3D

3D tooling where every asset carries its own proof.

Rupa3D builds 3D — exact b-rep CAD (OpenCascade), meshes (headless Blender), and scenes with physics (three.js + Rapier) — and then measures what it built against closed-form references or independent measurements. The measurements travel with the asset: a certificate embedded in the GLB at asset.extras.rupa3d, readable with a few lines of code and no Rupa3D installed.

It is designed to be driven by AI agents over MCP, and by CI over a command-line gate.

Bahasa Indonesia

Status — 1.6.2 · 297 tests · MIT · on npm as rupa3d · in the official MCP Registry as io.github.tiranyx/rupa3d. Developed and tested on Windows 11 with Node 22. Tool descriptions and messages are still in Indonesian.

Use it as an MCP server

claude mcp add rupa3d -- npx -y rupa3d

Or in any MCP client configuration:

{ "mcpServers": { "rupa3d": { "command": "npx", "args": ["-y", "rupa3d"] } } }

Requires Node ≥ 22.14. Workspaces live in ~/.rupa3d; override with RUPA3D_RUANG.

Blender is optional. 23 of the 36 tools run without it:

| without Blender | tools | |---|---| | Check any GLB against a spec; the certificate is embedded in a copy of the file | rupa_periksa with berkas | | Triangle topology, texel density (pixels per metre), VRAM per texture format | rupa_topologi, rupa_tekstur | | Convex collision hull, with how far it misses the shape | rupa_proksi | | Exact b-rep CAD: primitives, booleans, fillets, 2D sketch → solid, exact volume and area, STEP/IGES/STL/GLB export | rupa_cad_* (11 tools) | | three.js scenes with Rapier physics, published as a single HTML page | rupa_adegan_* (8 tools) |

The other 13 drive headless Blender — bpy modeling, import (GLB, OBJ, FBX, STL, PLY, SVG), renders, LOD, baking, collision proxies, source-mesh topology. Blender is found under C:\Program Files\Blender Foundation\*, at RUPA3D_BLENDER, or as blender on PATH.

What "proves" means

Every number below comes from the test suite or the documentation in this repository.

• Sketch → solid. Extrusion volume is checked against area × distance (area from the shoelace formula), revolution against Pappus's theorem. Recorded errors range from 0 % to 6.03e−14 %; the tests fail above 1e−10 % for extrusion and 1e−8 % for revolution. A result that misses its reference is never saved.
• Exact vs approximate. For a Ø140 mm flange the b-rep volume equals the analytic value; the best mesh tessellation tried (9,252 triangles) is off by −0.0016 %. CAD export reports which of the two you are holding (eksak) and by how much the approximation misses.
• The kernel's ok: true is not accepted as proof. A fillet larger than half the smallest edge came back ok with a broken solid whose volume grew by 1,971 mm³ on a 6,000 mm³ box. Independent checks reject it.
• Physics. Free-fall time matches h = ½gt² within one simulation step at 30, 60 and 240 Hz, and the error shrinks as the step shrinks — the test guards the direction, so it cannot pass on a lucky timestep.
• Compressed files. A Draco-compressed sphere receives the same certificate as its uncompressed twin: 2,208 triangles, 0 failures, 6 warnings.
• Hostile files. GLBs with node cycles, out-of-range indices, forged accessor counts or oversized skins are rejected with a message naming the defect, instead of hanging or passing. Each case runs in a child process with a time limit, so a regression fails the suite instead of freezing it.

Command-line gate

npx -p rupa3d rupa periksa model.glb

Exit code 0 — passes every required rule. 1 — fails a required rule (fix the asset). 2 — could not be checked (fix the tool or the input). --json for machine output, --spek for another spec, --tempel to write a certified copy.

Develop

npm install
npm test

297 tests run against real Blender, OpenCascade and Rapier — not mocks — and one suite talks to the server over real MCP stdio, including from a foreign working directory the way an MCP client launches it. Tests that need a local asset are skipped with the reason and how to obtain it; node contoh/flange-cad.mjs rebuilds the flange fixture in about 40 seconds without Blender.

Limitations

• Tool descriptions, messages and most documentation are in Indonesian.
• Tested on Windows 11 only. macOS and Linux are untested.
• rupa_skrip runs arbitrary bpy Python inside Blender. That is the modeling capability, not a bug — but do not expose the server to untrusted input over a network.
• Hardening against hostile files is partial: there are no general input size limits and no process isolation yet.
• It is not a replacement for Blender or Spline today.

License

MIT. Dependencies keep their own licenses — notably brepjs-opencascade (LGPL-2.1-only). The scripts in bpy/ run inside Blender.

Documentation (Indonesian)

| | | |---|---| | docs/INTEGRASI.md | using Rupa3D from outside: library, MCP server, and reading the certificate without installing anything | | docs/SERTIFIKAT.md | specs and certificates | | docs/BERKAS.md | checking other people's GLB files without Blender | | docs/CAD-MCP.md · docs/SKETSA.md · docs/TAHAP-B-BREP.md | b-rep CAD over MCP, and 2D sketch → solid | | docs/FISIKA.md · docs/ADEGAN.md | physics, scenes and the web runtime | | docs/TOPOLOGI.md · docs/TEKSTUR.md · docs/DRACO.md | topology, textures, compressed files | | docs/PERAKIT.md · docs/OPTIMASI.md | the one-call asset pipeline, and performance | | CHANGELOG.md | every release — including a Terbongkar ("exposed") section for what broke, measuring tools included |