Downloading and using the XML file as the source for some transformation would produce what is considered a derivative work, which would thereby be covered under the GPL. By accessing the file contents through an "abstract interface" this limitation could be avoided.
As I understand it, because the XML file contains instructions with regard to how and where to read files for the purpose of discovering their MIME type, those instructions are copyrightable. I could be wrong, as I am not a lawyer.