Geophysically detectible mid-lithospheric discontinuities (MLD) and lithosphere-asthenosphere boundaries (LAB) beneath cratons have received much attention over recent years, but a consensus on their origin has not yet emerged. Cratonic lithosphere composition and origin is peculiar due to its ultra-depletion during plume or accretionary tectonics, cool present-day geothermal gradients, compositional and rheological stratification and multiple metasomatic overprints. Bearing this in mind, we integrate current knowledge on the physical properties, chemical composition, mineralogy and fabric of cratonic mantle with experimental and thermodynamic constraints on the formation and migration of melts, both below and within cratonic lithosphere, in order to find petrologically viable explanations for cratonic mantle discontinuities. LABs characterised by strong seismic velocity gradients and increased conductivity require the presence of melts, which can form beneath intact cratonic roots reaching to ~ 200-250 km depth only in exceptionally warm and/or volatile-rich mantle, thus explaining the paucity of seismical LAB observations beneath cratons. When present, pervasive interaction of these - typically carbonated - melts with the deep lithosphere leads to densification and thermochemical erosion, which generates topography at the LAB and results in intermittent seismic LAB signals or conflicting seismic, petrologic and thermal LAB depths. In rare cases (e.g. Tanzanian craton), the tops of live melt percolation fronts may appear as MLDs and, after complete lithosphere rejuvenation, may be sites of future, shallower LABs (e.g. North China craton). Since intact cratons are presently tectonomagmatically quiescent, and since MLDs produce both positive and negative velocity gradients, in some cases with anisotropy, most MLDs may be best explained by accumulations (metasomes) of seismically slow minerals (pyroxenes, phlogopite, amphibole, carbonates) deposited during past magmatic-metasomatic activity, or fabric inherited from cratonisation. They may accumulate as layers at, or as subvertical veins above, the depth at which melt flow transitions from pervasive to focussed flow at the mechanical boundary layer, causing azimuthal and radial anisotropy. Thermodynamic calculations investigating the depth range in which small-volume melts can be produced relative to the field of phlogopite stability and the presence of MLDs show that phlogopite precipitates at various pressures as a function of age-dependent thermal state of the cratonic mantle, thus explaining variable MLD depths. Even if not directly observed, such metasomes have been shown to be important ingredients in small-volume volatile-rich melts typically penetrating cratonic lithospheres. The apparent sparseness of evidence for phlogopite-rich assemblages in the mantle xenolith record at geophysically imaged MLD depths, if not due to preferential disaggregation in the kimberlite or alteration, may relate to vagaries of both kimberlite and human sampling.