When you create an Agent Handler you can configure which existing agents should communicate with it from the 'Agent Handler' section of the ePO console. These are called Agent Handler Assignments.
You may want to consider the reasons behind why you are looking to implement ab Agent Handler. Typically Agent Handlers are deployed in large environments (10000+ nodes) to reduce bandwidth for communication to ePO server or they can be deployed in a DMZ to facilitate Agent-to-Server communication with roaming systems over the internet (remote laptops without vpn access to network).
As you mentioned updates would you not find distributed repositories a more useful implementation?
Alternatively if you are looking to route all traffic from the servers to the ePO server through one firewall rule you could consider using a Relay Agent (introduced by McAfee Agent 5.x)