We have compiled a sample of 234 ultra-steep-spectrum (USS) selected radio sources in order to find high-redshift radio galaxies. The sample covers the declination range -40{deg}<{delta}<-30{deg} in the overlap region between the 1400-MHz National Radio Astronomy Observatory (NRAO) Very Large Array (VLA) Sky Survey (NVSS), 408-MHz Revised Molonglo Reference Catalogue and the 843-MHz Sydney University Molonglo Sky Survey (the MRCR-SUMSS sample). This is the second in a series of papers on the MRCR-SUMSS sample, and here we present the K-band (2.2{mu}m) imaging of 173 of the sources primarily from the Magellan and the Anglo-Australian Telescopes. We detect a counterpart to the radio source in 93 per cent of the new K-band images which, along with previously published data, makes this the largest published sample of K-band counterparts to USS-selected radio galaxies. The location of the K-band identification has been compared to the features of the radio emission for the double sources. We find that the identification is most likely to lie near the mid-point of the radio lobes rather than closer to the brighter lobe, making the centroid a less likely place to find the optical counterpart. 79 per cent of the identifications are less than 1arcsec from the radio lobe axis. These results differ from studies of low-redshift radio samples where the environments are typically not nearly so dense and disturbed as those at high redshift. In contrast to some literature samples, we find that the majority of our sample shows no alignment between the near-infrared and radio axes. Several different morphologies of aligned structures are found and those that are aligned within 10{deg} are consistent with jet-induced star formation. The distribution and median value of the K-band magnitudes for the MRCR-SUMSS sample are found to be similar to several other USS-selected samples even though each sample has a very different median 1400MHz flux density. USS selection from a lower radio-frequency sample has not netted fainter K-band magnitudes, which may imply that the k-correction is not responsible for the effectiveness of USS selection.