Studies utilizing eye-tracking methods have potential to promptly capture real-world dynamics of one of the core areas of vulnerability in autism spectrum disorders (ASD), selective social attention. So far, no studies have successfully reported utilizing the method to examine social attention in toddlers with neurodevelopmental vulnerabilities in real world and challenging settings such as an interactive face-to-face. This study examined the feasibility and validity of live eye-tracking method in response to live interaction occurring in several contexts in toddlers with and without ASD. Forty-seven toddlers with ASD, with atypical development (ATYP), or typically developing (TD), underwent a 30-s live eye-tracking procedure during a face-to-face interaction with a masked stranger using child-directed-speech (16 ASD, 14 ATYP, 17 TD; M = 23.44 months, SD = 6.02). Out of this group of toddlers, 29 (10 ASD, 8 ATYP, 11 TD, M = 21.97 months, SD = 5.76) underwent the same procedure with one of their maskless parent. Task completion rate, calibration accuracy, and affective response (feasibility measures) as well as attention to the task and the social partner (validity measures) were examined. Task completion rate and calibration accuracy were excellent. Despite the challenging context of face-to-face interaction, the toddlers exhibited a neutral affect, and high attention to the task and the speaker. As anticipated, toddlers with ASD looked less at the social partner compared with control groups. However, attention was comparable between the Stranger and Parent conditions, indicating that the effect was consistent regardless of presence of face covering or the familiarity of the interactive partner. The study demonstrates the high feasibility and validity of a live eye-tracking task involving face-to-face interaction in neurodiverse toddlers with social vulnerabilities. The effect of diminished attention to social partners in toddlers with autism is robust and present when interacting with an unfamiliar person and parent. The results suggest that a brief live eye-tracking method constitutes a promising ecologically valid candidate biomarker and potential intervention outcome in autism.