Probe-based confocal laser endomicroscopy (pCLE) is a vital technology to realize <i>in-vivo</i> imaging of the organ surface in real-time. However, the field of view (FOV) of pCLE is too small to get a full picture of targeted tissues for diagnosing. It is recognized that collecting the image stream and assembling the images in a large mosaic can overcome FOV limitations. Many studies have been carried out to develop robust and fast methods for image processing to get a clear mosaic. However, it is still difficult to obtain high-quality images consistently due to the presence of tissue deformations on soft tissue surfaces. The control of probe movement is still a key research challenge. In this work, a position-based visual servoing scheme is proposed to achieve high accuracy and gap-less mosaics for soft tissue endomicroscopy. Experimental results show that the proposed control scheme with visual servoing can obtain a large and accurate spiral mosaic (approximately 7 <i>mm</i><sup>2</sup>) without gaps. Compared to open-loop control, the obtained trajectory better matches with the target one, with the error of circular trajectory reduced by 31.79% and the error of spiral trajectory reduced by 46.86% on average. The proposed control scheme has a high potential to achieve a large and accurate mosaic for soft tissue endomicroscopy.