Anbuk, as per some of my other responses, you really should be discussing this question with whomever you purchased your Juniper equipment from, or additionally your local Juniper account team.
MC-LAG and [true] Spine-Leaf designs are 2 different choices of architecture - you can NOT do both at same time, IMHO. MC-LAG requires a connection between what would be the equivalent of 2 "Spine" switches. MC-LAG can not expand beyond to "Spine/Core" switches. To increase number of Core switches with MC-LAG, you need to create 2 separate MC-LAG networks, and then interconnect them. The interconnect could also be MC-LAG, or other technologies.
In [true] Spine-Leaf there is no interconnection between Spines and depending upon the technology of choice, the number of Spines is potentially unlimited. Arul mentioned VCF, which is basically an extension of VC that can scale to greater number of switches (10 max for VC; 20+ max for VCF) but uses a Spine-Leaf architecture as well. Both VC and VCF support a mixture of both L2 and L3 as required.
Now any network tolopology can be made into an IP-Clos although Spine-Leaf (non-VCF) is generally the prefered method. IP-Clos requires a pure L3 IP underlay, with iBGP or eBGP generally the protocol of choice. These networks then generally use some form of over-lay if L2 communication (for something like vMotion) is required. Some L2 technology that can tunnel across the L3 underlay is the solution. VXLAN is one of the most common over-lay currently being used.
There are multiple choices for network architecture, and you should discuss this with your Juniper reseller or Juniper sales team as to which one might be the best for you, as there are pros and cons to every approach.
Good luck.