Adequate grip ability is important for effective execution of daily living activities. Neurological disorders like stroke that result in muscle weakness, limited strength and poor control often lead to reduced grip ability in the affected limb. Conventional rehabilitation for grip training is often monotonous and subjective. Technology-assisted Virtual Reality (VR)-based rehabilitation offers a motivating environment to the participants for rehabilitation. However, the existing systems need specialized set-up architecture, thereby limiting their accessibility. Furthermore, these systems quantify the functional grip ability based on task performance, and do not explore physiological basis of grip ability. In this work, we develop a VR-based rehabilitation platform integrated with physiology-sensitive biofeedback. The developed platform, Gripx makes use of features extracted from sEMG data collected from upper limb muscles to adaptively provide audio-visual biofeedback through a VR environment. We compare task based performance, physiological indices and clinical measures to evaluate the potential of Gripx. The results of our study with 8 healthy and 12 post-stroke participants show the potential of Gripx to contribute to grip rehabilitation over multiple exposures. This approach of integrating VR-based task design with physiology-sensitive biofeedback helps patients to better assess their physiological responses and enhance the efficacy of rehabilitation.