In this paper, the non-zero finite positive state consensus problem of leader-following multi-agent systems (MASs) connected over a directed network is investigated. The multi-agent network consists of homogeneous linear time-invariant (LTI) positive agents whose minimal positive state-space realization is assumed to be known. In this paper, a state-feedback hierarchical control protocol is proposed where the local controller synthesizes a singular, Lyapunov stable, and Metzler system matrix with a simple dominant eigenvalue at origin. The obtained consensus vector is the positive eigenvector associated with the zero eigenvalue of the synthesized system matrix. The controller gain matrices are obtained by solving a set of necessary and sufficient conditions derived in linear programing framework. A numerical example is given to elucidate the usefulness of the proposed results.