In this technical note, a descriptor approach to leaderfollowing output consensus of multi-agent systems with both stationary and dynamic leaders is given in the presence of transmission delay and model uncertainty. The proposed method can deal with stable and unstable agents described by general linear models. To this end, a new proportional-derivative- integral (PID) consensus protocol for the closed-loop multi-agent system is proposed under a directed graph. Applying this consensus protocol to the multi-agent system leads to a time-delay closed-loop equation of neutral type. To deal with the resulting neutral system, a descriptor model transformation is used to derive delay-dependent sufficient conditions for the existence of the consensus protocol in terms of certain linear matrix inequalities (LMI). The application of the proposed method is illustrated in a teleoperation system. Simulation results are given to show the effectiveness of the proposed approach.