Using a different port in itself should not really make a different from the agent's point of view, so that's fine.
So in which application we need to use that manual transaction code?
This basically depends on what the agent captures out of the box - worst case the agent won't capture anything, in that case you'd need to manually instrument the code both in the client and in the server side. If you see part of the request showing up in Kibana out of the box - i.e. you see the server side and you don't see the client side which calls into the WCF service (this is a likely scenario) then you'd need to instrument the caller code.