I am not aware of such functionality in existing online repositories at the present time. However, depending on your needs, skills and willingness to dedicate your time and efforts to such project, you can consider writing custom software, based on natural language processing (NLP) approaches, in particular information extraction, for parsing papers of interest - individually or in bulk - in order to extract their reference list information. I believe that it should be relatively easy to write such software, as the semantic complexity of reference list sections is relatively low. Having said that, I see some potential difficulties, related to the quality (consistency) of listed references in terms of content, formatting and publication style.
I think that using such software would be perfectly legal and ethical, as that part within each paper doesn't really represent an intellectual property or is copyrighted (please correct me, if I'm wrong).