Conventional endoscope-guided photodynamic therapy (PDT) suffers mostly from motion artifacts, therefore expert hand-eye coordination was always needed during manual operations. In this paper we introduced a visual servo scheme to handle the tracking problem between the focused area and the targeted lesions. The scheme is consisted of real-time feature matching, relative motion cancellation and real-time light dose surveillance. Experiments were carried out both on simulated data and a silicon phantom. It indicates that this scheme outperforms the conventional scheme in terms of reduction in operation time and exposure to healthy tissue.