One (for sure) or even three (possible) header files contained in the MGH/MF Waveform Database are faulty and therefore can't be read with the wfdb python library
mgh025's header is missing a digit in its year (e.g. 199 instead of 1991?)
mgh061's as well as mgh127's headers are missing a correct declaration of the number of samples per signal field in their record line. E.g. in both headers the signal has a length of 0 samples - which of course can be a correct declaration but should there really be files in an official database missing a value in that field? [furthermore/alternatively could be an incompatibility of the python package]
One (for sure) or even three (possible) header files contained in the MGH/MF Waveform Database are faulty and therefore can't be read with the wfdb python library