In that case, you need to find out from your MPLS vendor what, if any, QoS support they provide. Usually they offer some traffic class structure that uses DSCP tags to identify traffic. Sometimes there's extra cost for QoS support or cost for different kinds of QoS support. Since you mention voice, you'll be expecially interested in what real-time traffic QoS support is available (this too sometimes has a separate charge).
Assuming "typical" MPLS provider QoS support, four (or more) traffic classes, with a real-time traffic class, you'll classify and mark you're traffic as you pass it off to your provider. Additionally, if you manage the actual CE device, you also organize your queues, and their bandwidth allocations, to correspond with the MPLS provider's QoS model. (NB: strictly speaking, you can model ingress into the MPLS cloud differently from what the provider expects, but's it's crucial you properly tag traffic to utilize the provider's QoS template.)
On a CE router, you might have some CBWFQ policy like this:
policy-map aSample
class VoIP
priority percent 30 (often depends on contracted bandwidth per link)
set ip dscp EF
class busCrit
bandwidth remaining percent 40
set ip dscp AF21
class bulkyStuff
bandwidth remaining percent 5
set ip dscp AF11
class class-default
bandwidth remaining percent 15
set ip dscp 0 (or BE?)
Again, bandwidth percentage would normally correspond to MPLS provider's QoS model. More than 4 classes might be available, depends on provider and what they offer.
The two usual congestion points are your CE interface into the MPLS cloud and the PE interface exiting the MPLS cloud. The former you might control, the latter is generally always controlled by the MPLS provider.
How to classify traffic depends on your service requirements and, also again, what the MPLS provider can deliver for QoS.
MPLS often provides, besides different QoS classes, contracted bandwidth which might be less than link bandwidth. Such contracted bandwidth often interacts with their QoS template. Since there's also often a charge for contracted bandwidth, understanding what it really guarantees, and how it interacts with their QoS model, can allow you to "right size" it.
BTW, a correct QoS implementation, often can go far in avoiding over provisioning of WAN bandwidth (and minimizing WAN costs). Keep in mind, TelCos aren't often interested in you minimizing your WAN costs. I.e. if QoS and MPLS together are new to you, and assuming you haven't already contracted for some services, you might want to retain some consultation in how best to meet your service needs.
