Embodied intelligence could grant soft, robotic manipulators with increased performance in adaptability and dexterity. However, we currently have a lack of tools for the forward design of finger behaviours with high-dimensional morphology. We take an open-source parametric hand design, which mimicks the degrees of freedom and tendon routing of a human hand, as the basis for morphology and behaviour exploration. Then, through simplified modeling and behaviour classification, we explore 60000 simulated designs, verify 10 fingers with rapid manufacturing, and compare 3 complete anthropomorphic hands which leverage selected finger designs. For initial exploration of desireable behaviours, we evaluate two optimizations, reaction behaviour diversity and dynamic stiffness range, leading to real-world grasping force improvements and bias towards different manipulation behaviours such as pinching or tactile exploration. Overall, the kinematic model, behaviour analysis, and rapid real world testing through parametric design offers a promising platform for furthering our understanding of designing exploitable and adaptive behaviour during dexterous manipulation.