In this paper, we present a method for micro-endoscopic topography measurement utilizing two-wavelength holography. Initially, we evaluate the inter-core dispersion and cross-talk of two commercially available imaging fiber bundles (CFBs) and introduce the concept of virtual surface roughness as a limiting factor of achievable measurement resolution. Subsequently, we describe a micro-endoscope setup incorporating 3D-printed micro-optics, resulting in a total diameter of less than 450 µm. We evaluate the measurement accuracy using a pyramid-shaped test object and demonstrate that a relative measurement error of 7.5% can be achieved with a simple phase unwrapping approach. Moreover, we demonstrate how leveraging a deep learning approach from existing literature, tailored for heavily noisy phase maps, effectively reduces the relative measurement error. The standard deviation of the measurement error is 4.2 times lower with the deep learning approach.