Using BCI technologies in neural rehabilitation, we can substantially improve hundreds of lives of people after stroke with more effective rehabilitation in restoration of their motor control. In this study, we demonstrate a neuro-rehabilitation framework which integrates BCI and a robot device to provide an active upper limb physical therapy. Our target population for neuro-rehabilitation are patients with cerebro-vascular brain damage and who lost their motor control on their upper limbs. We analyze brain hemodynamic responses as a neural features associated with a cognitive task and decoding is performed using support vector machine classifiers. The decoded commands from neural signals are sent in real-time to control a phantom haptic device into desired two directions. The experimental setup is validated with seven healthy subjects and the obtained empirical results shows the feasibility of this approach for stroke rehabilitation purposes.