Since most clustering algorithms operate from a pariwise distance or similarity matrix, you need to compute one such matrix that captures relevant similarities between the proteins in your list. Molecular functionality is somewhat imprecise. I generally take it to mean what's covered by the molecular function domain of the Gene Ontology which you can use to compute some similarity between proteins, for example with the R Bioconductor package GOSemSim.