We present results from our research into assistive robots for highly unstructured environments. The concrete type of robot we studied is a robotic scrub nurse (RSN) for microsurgery. In microsurgery the surgeon works behind a microscope and must maintain field of view as much as possible. The task of managing and delivering instruments to the surgeon rests with the scrub nurse. During our attempt to design a robot which could automate this delicate but repetitive task we have identified the robot interface to instruments and human subjects as the essential component and have made that the central topic of our work. Special attention was paid to the robot-human interface at the haptic level. To that end we designed and evaluated a shape conforming grasp mechanism to act as a soft touch interface constantly adapting to the human touch. Such approach is uncommon in the robotic industry where per-task adapter-style exchangeable grasping mechanisms are mounted on the robot for each run. While industrial robots deal with few instruments and must lift heavy weights, this problem domain contains hundreds of instruments and prohibits any type of crushing or cutting movements. Our results show that the application of universal soft touch grippers is justified in environments where humans interface robots.