Measuring human arm stiffness in robot interaction is a crucial topic in both neuroscience and the related learning process during skill acquisition and functional recovery in neurological subjects. However, it is a complex and time-consuming procedure often requiring a computational burden which prevents from an online estimation of the data. Most systems described in the previous literature uses robotic manipulandum to estimate limb stiffness by perturbing the arm across different directions over multiple trials and acquiring the corresponding restoring forces. The proposed method is still robust and accurate, although with rather strong limitations in terms of speed and acquisition bandwidth. For this reason, we designed a mechatronic device able to carry out endpoint stiffness estimation within in a single trial. The proposed system can be operated in a stand-alone configuration or can be plugged in a robotic manipulandum, allowing us to perform the measurement during a posture maintaining task in contact with the robotic counterpart. This paper describes the mechanism and the design, testing the device in different experimental contexts, using a customized test bench to characterize the potentials and the limits of the proposed architecture. Furthermore, we tested the system on human subjects to obtain a reliable bidimensional estimation of arm stiffness when it is plugged in a robotic device. Results are reported and discussed in detail highlighting the limitations and the advantages of using the proposed solution.