MS is a clinical diagnosis: that means that the neurologist needs to take everything into consideration and make a diagnosis based on what he/she and other medical professionals can see by direct observation.
The McDonald criteria, just like all diagnostic criteria and tests are not 100% accurate. It's sort of an 80/20 rule. They try and work out the simplest, most effective criteria that capture as many people as possible that they can, but it will never be perfect. That means that sometimes people meet the criteria, but they don't have MS, and vice versa.
The McDonald criteria also stipulate that MS is a clinical diagnosis - that MRI isn't even necessary in some cases. However, if it's done and it's clear, then the neuro should be careful.
Anyway, the upshot is that if a patient presents with clinical exam results, symptoms and history that is wholly consistent with MS, and alternative diagnoses have been ruled out, then the neuro can diagnose MS without MRI or LP or anything. Of course, these days, everyone gets scanned. If the scan comes back negative or inconsistent with MS, then the neuro would have to be very confident that only MS could explain the patient's case before he/she can then diagnosis MS.
Most importantly, not everyone with MS has a positive MRI and MRI scanners are not perfect. Any neuro who relies on MRI is rubbish at his/her job: clinical signs, symptoms and history are FAR more important!