This work presents the proposal of a complete system for upper- and lower-limb rehabilitation for post-stroke patients, which uses a Brain-Computer Interface (BCI) based on Motor Imagery (MI), a robotic glove, and a robotic monocycle to recover both upper- and lower-limb movements. Common Spatial Pattern (CSP), Filter-Bank CSP (FBCSP), and Riemannian Geometry (RG) are used as features to be inputted to a classifier (Linear Discriminant Analysis - LDA). Preliminary experiments were conducted with two healthy volunteers and two post-stroke patients to detect their MI of pedaling and opening/closing the hand, reaching a mean Accuracy (ACC) of 80%, based on the analysis of their topographic maps of EEG signals, from mu (8–12 Hz) to beta (18–24 Hz) bands, acquired from sixteen electrodes located on: FP1, FP2, F3, F4, FC3, FCz, FC4, C5, C3, C1, C2, C4, C6, CP3, CPz, CP4, including two references at earlobes (A1 and A2). Results also show that it is possible to differentiate the MI of opening and closing the hand, which is essential for developing new neurorehabilitation protocols.