Today, therapy of neurologically induced functional losses of the upper limbs is mainly carried out manually. Recently, with the progress in automation technology, device-assisted therapy has established itself internationally as an additional treatment option. Thereby, proven therapy methods are applied automatically, which can relieve therapists, increase treatment frequency and lead to a better outcome. However, many of these solutions are still costly, heavy, bulky, or unsuitable for non-specialists because of the device's and external actuator unit's complexity. These limitations deny regular and otherwise immensely beneficial self-training by demanding the continuous presence of a therapist for setup and oversight. In this context, Shape-Memory-Alloy-based actuators' usage may permit new design approaches with enhanced physical properties and usability. Thereby, inhibition thresholds are overcome by reducing the size and weight of available devices and their peripherals, which use standard actuators. For SMA, the necessary strokes and forces are a challenge. To meet the given demands of automated grasp therapy, a suitable actuator build is designed using VDI 2248. The build bases on an antagonistic SMA approach consisting of Nitinol spring combinations to match given boundary conditions, like necessary stroke and gripping forces for a physiologically correct hand movement. Furthermore, optimization for properties like a minimal size to stroke ratio is conducted. This paper delivers an early proof of concept based on a prototype for an SMA-actuated grasp therapy device for neurological rehabilitation.