The services provided by learning object repositories are usually enabled by the processing of the metadata documents associated with these learning objects. This paper proposes a way to process these metadata documents, which are usually encoded in XML, through a framework called XLOP. XLOP is based on attribute grammars, a well-known technique used in the development of language processors. XLOP makes the automatic generation of efficient XML processing components from high-level specifications possible, enhancing the maintainability of the aforementioned services. The technique is illustrated in the context of Chasqui, a system for building repositories of learning objects in specialized domains.