In this brief, the problem of output consensus in a multi-agent system is considered, where a distributed dynamic output feedback control architecture is proposed. The control architecture consists of two components, one is local controllers for the agents, and another is a gain for the network. The transfer function of local controller is designed based on the number of positive real axis (in the complex plane) zeros of the agent transfer function, and then network gain is selected by sketching the root-locus of a unity feedback system. It is shown that the output consensus can be achieved with an integral type controller for a class of agents, irrespective of their order. Moreover, with the proposed control architecture, there is no need of communication between the local controllers. The effectiveness of the developed approach is demonstrated through numerical examples.