model a system with finite users as a Markov Chain

Question

I have to model a system M/M/2 with finite users (4 users) as a Markov Chain (and then find the probality an incoming users would enter the queue being the servers busy but that's not the problem). I am a little confused if a system with finite users can be modelled this way. The M says that the arrival process is markovian but how it can be? For example the state of the system is taken in consideration (if there are already 4 users in the system the arrival rate is 0) so the property that a markovian process must be "without memory" is invalid. Moreover the fact that the arrival rate when there are 4 users capable of arriving is the same as when there are only 2 users available is counterintuitive. Could someone explain please? Thanks