My preferred method is to keep the MX in NAT mode, and plug it into the Internet beside the existing firewall. I usually plug WAN1 into the primary Internet circuit, and then get the customer to get a cheap domestic fibre and plug it into WAN2. It's cheap insurance for a primary circuit failure.
If I can't plug WAN1 in directly to the Internet then I try to get a stub interface created between the existing firewall and the MX, and WAN1 goes there. Then I can still use NAT mode, two WAN ports and SD-WAN.
In each case, the inside of the MX goes to the inside of the network.
If I can't do any of that then in desperation I use the MX in VPN concentrator mode.
I also allocate a prefix (a supernet) for all the remote subnets and use static routing. The existing firewall default gateway should only need this single route then. I avoid dynamic routing if possible. It adds complexity that can often be avoided by design.
You sub-allocate your remote sites out of the supernet prefix.
If you have to run the MX through another firewall then I use a manual NAT forward for AutoVPN.
https://documentation.meraki.com/MX/Site-to-site_VPN/Automatic_NAT_Traversal_for_Auto_VPN_Tunneling_...
This works the most reliably and is the fastest of the AutoVPN rebuild options if there is a failure.