Hi,
If you are using cutout images from the Legacy Survey sky viewer, then (if you request cutouts at 0.262"/pixel) those are in nanomaggy units. Those images are coadds from many input images, so there is not a single conversion back to counts -- it's a different conversion for each image that went into the coadd.
What are you trying to do? What do you mean by MGE -- multi-gaussian expansion? My guess is that what you actually want is an uncertainty estimate on each pixel's flux estimate. If you add the "&invvar" argument to the cutout URL, you will get a multi-HDU FITS image with inverse-variances for each pixel, eg
cheers,
dustin