A 560k-view video got 800 stars: CNC Kitchen's ScanRuler, and the half you can copy

German 3D printing creator Stefan Hermann (YouTube channel CNC Kitchen, 767k subscribers) built a free 3D scan measurement and reverse-engineering tool that runs in the browser, and it lets Claude rebuild scans into editable CAD models on its own. Ten days after its public release it had only 90 stars; a week after one video it passed 900. The video drove the stars, and you can't borrow that. What you can copy is the niche he picked and the step of shipping it as an MCP server.

A fridge bracket he promised his aunt a year ago

In Oct 31, CNC Kitchen posted the video "3D Scanning Will Never Be the Same Again!" and published a written version on the blog the same day. It includes a small moment: Stefan's fridge bracket broke, and "a year ago I promised my aunt I'd fix it, but never found time to reverse-engineer it." This time he put the scan files in a folder, pinned a few notes on the broken faces, and let Claude take over. Claude aligned the scan, fit the geometric features, sketched, and extruded, producing a parametric model that could be edited directly, and it even filled in the broken section. He exported a STEP file and 3D printed it.

The same bit repeats through the video: for a cracked Vespa engine case, he modeled a patch wall and used a tool called "Trim to Scan" to trim it along the break, which he calls "virtual kintsugi." For a speedometer, Claude first reverse-engineered the shape, then designed a TPU protective case to spec, and the printed case fit.

The video was sponsored by scanner maker Elegoo and promotes its $1,599 new scanner, the M1 Ultra, labeled "includes paid promotion." As of 11-06, it had 567k views, 16k likes, and over 1,200 comments (YouTube page, 11-06). Every demo project in the video links to scanruler.com, which you can open directly.

What it is

scanruler.com works as soon as you open it: no install, no sign-up, no cookie banner, and files are processed only in your own browser. The code is on GitHub (CNCKitchen/ScanRuler, TypeScript, AGPL-3.0, with a README noting you can buy a commercial license).

Six workspaces:

• 3D measurement: click a point on a scan to auto-fit spheres, cylinders, cones, and planes, then add dimensions, tolerances, and GD&T checks, modeled on industrial software GOM Inspect and PC-DMIS;
• Deviation: auto-aligns a scan to the original CAD and outputs a color deviation map;
• Wall thickness: one click for a wall-thickness color map;
• 2D measurement: uses a flatbed scanner as a measuring microscope, exports SVG / DXF;
• Reverse engineering: sketch, extrude, revolve, and fillet on a scan, output STEP solids;
• Mesh tools: hole filling, decimation, boolean operations.

Add an MCP server (npx scanruler-mcp), and Claude Code, Codex, and Cursor can all drive the page open in the browser. The model runs on the user's own subscription, so the author pays zero inference cost. Stefan writes in the blog that Claude Opus 5.5 gives the best results for now, weaker models "sometimes do a terrible job," and local models didn't work on his machine.

He says plainly it was built with vibe coding: "I know not everyone likes this approach... for me, it's a tool to build the tools I've always wanted." It took a few weeks to finish. By 11-04 there were 146 commits, and 11-01 shipped two releases in one day, 0.9.3 and 0.9.4.

When the stars came

Data: GitHub daily stars / forks (from 10-28), npm downloads API (read 11-06). GitHub live count as of 11-06 was 952.

The table says it plainly: same product, industry press coverage on launch day, and just under 100 stars in more than a week. The video went out and within 48 hours it jumped by 430. After 11-04 it settled back to +50 to +100 a day.

One more number to remember: 567k views brought in only about 840 stars, under 0.2%. Most CNC Kitchen viewers are people who do 3D printing, not GitHub users. They want a tool that works, not to click a star. For anyone hoping to build stars, that means the star ceiling in this niche is low to begin with.

On the other side, there are 162 forks, about 17% of the star count, which is higher than typical for popular projects. People have already made a Chinese version (scanruler-zh), a Russian version, and standalone builds, and each sits at 0 to 1 stars. People really are using and modifying it, but "ship a localized fork" doesn't earn stars on its own.

Why this product

A free channel that got shut down by an acquisition. The 3D printing crowd used to test scans with GOM Inspect's free version. After Zeiss bought GOM and renamed it ZEISS INSPECT, Stefan says the free version required a company email to use. The remaining Geomagic and PolyWorks cost five figures in dollars per seat, and even QuickSurface's cheapest tier runs about €480 a year. Meanwhile consumer scanners keep getting cheaper and more accurate. He opens the blog with: "A lot of people's 3D scanners ended up as expensive paperweights. The scanner isn't the problem, the software is." The demand was already there, and so was the "free alternative" pitch.

Credible. The README benchmarks against a baseball bat (two precision balls on a carbon fiber rod): ScanRuler measures the ball center distance as 148.639 mm, GOM Inspect as 148.64 mm, matching down to GOM's last displayed digit. Stefan is an engineer, and his channel is known for rigorous testing. For a measurement tool, that counts for more than any feature list.

The AI angle is strong. "Scan the broken part, AI gives you back an editable CAD model" grabs attention as a video or a screenshot. On X, users fed a thermos scan to Claude and got an AirTag holder (@jmedlin); someone recreated a zinc alloy casting with GPT (@LieutenantRusty); there were reposts in Japanese, Chinese, and Polish. 11-03 Another channel, Making for Motorsport, posted "Vibe-Modeling is here," modeling directly from scans with ChatGPT, with 240k views. The "AI + scan reverse engineering" topic is hot on its own.

A sponsor paid the promotion. Scanner makers worry most about people buying a scanner and not knowing what to do with it. A free, useful app fills that gap perfectly, so Elegoo was happy to sponsor. Stefan built the video as a combined demo of "the scanner + my free software," with the scanner's affiliate link in the description. This was the smartest move in the whole thing.

Who's competing for this

The 3D Printing Journal says it's the third tool in this space in three months. Free, no install, runs locally, industrial-grade measurement methods, agent-ready. It's the only one that checks all of those boxes right now.

Can you follow it?

Cross off what you can't borrow: a 767k-subscriber channel, a scanner maker willing to pay you, years of engineering reputation banked. Those three together are what carried the jump from 90 to 900. Build the same thing head-on in 3D scanning with none of that, and you'll likely end up where it launched: a few press write-ups and a few dozen stars.}

What you can copy is the method, and you don't need to use it on 3D scanning:

1. Find a "free version that got shut down after an acquisition." When a big company buys a pro tool and puts up a paywall on the free version, the old users are fed up, and the "free open-source alternative" pitch spreads on its own. Start by searching forums and Reddit for "XX free version is gone."
2. Make it run in the browser. No install, no sign-up, no file uploads. People in this niche care about that a lot.
3. Add an MCP so users bring their own model. You pay no inference costs, and you get shared in the "AI does the pro work for me" conversation. What's really new here isn't MCP itself. It's giving agents a dedicated toolset (fitting, measurement, deviation checks) instead of making a general model chew on raw triangles.
4. Check your output against a verifiable number from an established tool. For something like 148.639 vs 148.64, one line is enough.
5. Tie the launch to an audience. If you don't have a channel, find a vertical creator who has an audience but lacks content, or a hardware vendor that needs a software ecosystem. The first needs demo footage, the second needs "it works with what you bought."

It also leaves gaps of its own. Stefan says on the blog that more complex parts, freeform surfaces, and rough castings often give "not very good" results. The forks people are making, for internationalization and standalone packaging, show that some users want an offline version and their own language. But those are PR-sized jobs for the original project, and a standalone fork alone won't pull many stars (inference).

If it were you

C: Worth reading, but building a scanner tool off this one isn't worth it. Its stars came from the author's own channel and sponsors, which you can't replicate; 3D 3D-printing viewers don't star much either, and 560k views only turned into 800 stars. The playbook is what's worth taking: "free alternative after an acquisition + runs locally in the browser + MCP + one accuracy number that holds up."

• Who it's for: Developers who know a specialist software community well (CAD/CAM, PCB, metrology, laser cutting, microscopy imaging, geospatial mapping), who know a free version in that world recently got pulled or paywalled, or who know a creator or hardware vendor in that field.
• Not for: anyone who wants to build a head-on competitor in 3D scan reverse engineering (Stefan iterates fast, shipping two versions a day, and the same direction already has Backflip, OwlCAD, and SolidCraft); anyone who only wants a localized or packaged fork (the existing ones are at 0 to 1 stars); and anyone with no distribution channel who's counting on natural GitHub traffic.
• How to do it: Pick a community, first confirm that "the free option is gone or now requires sign-up" is actually true, then build the core features to run in the browser, add an MCP and an operating guide for agents; put a result that matches an established tool digit for digit on the first screen of the README.
• How to get users: Before launch, turn the demo project into a clickable link and give it to a YouTube creator in that community, or negotiate a joint video with a hardware vendor that needs companion software; also post in the relevant r/ subreddits and to industry media. Don't rely on HN alone: ScanRuler's data shows industry media coverage only brought in a few dozen stars in a week.
• Window: before the trend hits: "AI agent does the work inside professional software" is hot right now (11-03 one ChatGPT scanning-and-modeling video drew 240k views). Expect copycats in every professional niche within about six months. The earlier you become the first free tool that plugs in an agent in a given niche, the better.
• What would mean I read it wrong: (1) ScanRuler gains another 3,000 stars within a month after the video buzz fades, which would show this GitHub crowd is bigger than I assumed and a direct 3D scanning or reverse-engineering tool deserves a second look; (2) Zeiss re-opens a free tier, or a scanner vendor builds the same measurement and agent features into its bundled software, and the "free alternative" angle is gone; (3) someone with no channel and no audience, just community posts, ships an MCP version of a similar pro tool and gets past a thousand stars, which would mean distribution matters less than I think and this opportunity should move up a tier.