Complex docs in CONTENTdm seem to have their own item records. See...
[http://reflections.mndigital.org/cdm4/item_viewer.php?CISOROOT=/mpls&CISOPTR=2881][1]
...for example. This might be a reasonable place from which to grab an MDL id. It comes close to making a good place for a landing from a comment page.
Note that this page has a link (Access this item) that provides an XML document with all the pointers of the pages...
[http://reflections.mndigital.org/cgi-bin/showfile.exe?CISOROOT=/mpls&CISOPTR=2881&filename=2547.cpd][2]
The link to the whole document would be to...
[http://reflections.mndigital.org/cdm4/document.php?CISOROOT=/mpls&CISOPTR=2881][3]
I think this means that I could sense an incoming commenter from a complex document by looking for "document.php" in the referrer. Then I'd have to modify my link back to Reflections based on whether I see "document.php" or "item_viewer.php". (Note, even if I see "item_viewer.php" I should maybe be on the alert for the viewer possibly coming from a complex document item?)
In either case, the screen scraping could be done from the "item_viewer.php" page, and if the MDL id appears there we are golden.