Regaining fine motor skills (FMSs; that is, the ability to make precise and coordinated movements of fingers) is the ultimate goal of hand rehabilitation. Although robotic-assisted technologies have been widely explored to help patients with simple hand activities, existing rehabilitation gloves still lack sensory feedback for closed-loop control or involve bulky external hardware, incapable of providing precise FMSs rehabilitation portably. Here we develop a soft rehabilitation glove that can accomplish diverse FMSs by integrating 15 bending sensors and 10 shape-memory-alloy (SMA) actuators. Three joint angles of each finger can be precisely sensed and thus are fed to the control system to actuate SMAs in a closed-loop manner. A touchable human–machine interface is also integrated to provide facile interaction for multi-modal rehabilitation exercises. Weighing only 0.49 kg, our soft glove has high portability, allowing for repetitive hand rehabilitation as needed. We validate that our glove can assist an individual with hand impairments after a stroke to realize a set of single and complex FMS rehabilitation exercises and to complete some activities of daily living.