Technological advances are increasing the range of applications for artificial intelligence, especially through its embodiment within humanoid robotics platforms. This promotes the development of novel systems for automated screening of neurological conditions to assist the clinical practitioners in the detection of early signs of mild cognitive impairments. This article presents the implementation and the experimental validation of the first robotic system for cognitive assessment, based on one of the most popular platforms for social robotics, Softbank “Pepper”, which administers and records a set of multi-modal interactive tasks to engage the user cognitive abilities. The robot intelligence is programmed using the state-of-the-art IBM Watson AI Cloud services, which provide the necessary capabilities for improving the social interaction and scoring the tests. The system has been tested by healthy adults (N = 35) and we found a significant correlation between the automated scoring and one of the most widely used Paper-and-Pencil tests. We conclude that the system can be considered as a screening instrument for cognitive assessment.