This is the opposite of what I have found. I wouldn't use 3D printed parts for any sort of interfacing / tolerance check because pla usually prints > 0.2mm out of spec.
But, maybe you have a better 3d printer or I misunderstood exactly how you use it.
Do you have a 3d printer that can made dimensionally accurate parts? Or would you mind clarifying the use case?
Perhaps gear meshing was a bad example requiring higher tolerances. You'd not be able to set the depth to a tolerance of a few microns, but you'd know if you were off by a factor of 2 due to confusion calculating diametral pitch vs OD.
We most recently used it to check timing belt compatibility, whether the belt could be installed over the pulley using the very limited travel available in the belt tensioner, or whether we needed a pulley with smaller flanges or a different way to adjust and change belts. Trying to figure out in CAD whether you could stretch a belt over the pulley, or slip the pulley over the shaft while the belt was pre-installed, is pretty difficult. Printing it and proving that it was easy cost $2 and reduced risk a lot.
With the most fine settings the slicer will do, it's within 20 microns in all dimensions, and usually it's so accurate my calipers show no error at all, which puts the error either under 5 microns or 1 micron, I forget.
The biggest issue with PLA is it shrinks a bit, so a perfect print is usually about 102%ish of the drawing size; the slicer manages all of that, though.
*People who don't work in CAD may not get it, but threaded interfaces are basically modeled with cylinders that overlap where the threads are engaged. It's funny what gets represented accurately and what doesn't, but just remember that the map is not the territory!
For weird things? Full parametric models would be great!
As an industrial engineer, most of my vendors offer this. I can get 3D CAD for robot arms, grippers, IO blocks, motors, cameras, enclosures, valves, cylinders, and just about everything else that goes in my machines. However, you're right that these don't have the same scale that McMaster does; they're selling the huge variety of nuts and bolts that bring these separate vendors together.
One of the underrated parts of this kind of CAD offering is that you can test out 3D printed parts directly from their CAD. Not sure if you're accurately understanding their otherwise very helpful documentation on the gear modulus of that thousand dollar hardened, ground steel rack-and-pinion? Download the step file and print it in a few bucks of PLA. No, it won't actually be strong enough to move that 3-ton casting even once, but you can trivially check that the meshing works as you expect.