Record in layers, so that you have flexibility when mixing. I've recorded urban ambiences at street level, on a 3rd story, and on an 8th story rooftop, mics pointing down and up, and it all sounds incredibly different. A mix of perspectives, times of day, and amount of traffic (vehicular and/or human) will all make a huge difference. To get really usable layers, it's harder and takes more time than you'd think.
If you do it right, IMO, recording urban ambiences is pretty much the same as shooting natural ambiences: Scout locations, listen keenly, think about final use, record longer than you'll need by a factor of 2x-3x, and determine per location if a spot sound or the broader scene is what you want to capture. Biggest difference, though, is that you don't really need to worry about mic self-noise in a city the way you do for rural recording.
I like ORTF for the widest and most enveloping ambiences, but truth be told, for portability and convenience I usually roll with an M/S rig. This has saved my bacon loads of times as I try out a location ambience session and find that actually that truck/streetsweeper/car/motorcycle/crazyman is a far more interesting spot source, so I switch focus in the field and then break it down as a mono layer in post.
I often just run experiments where I build urban ambiences in post against specific emotion keywords: Stressful, intimidating, bucolic, etc. Good practice to see where the holes in your library exist!