Wireless Capsule Endoscopy (WCE) is an innovative technology in the medical field for the diagnosis of the Gastrointestinal (GI) tract. However, the inherent hardware limitations of WCE sensors resulting in low spatial resolution of the acquired images impair diagnostic accuracy. Super-Resolution (SR) techniques are used to enhance details in images and to reconstruct High-Resolution (HR) data from given Low-Resolution (LR) image. The state-of-the-art deep learning-based SR methods often rely on supervised training, where LR images are typically synthesized by applying known degradation techniques, such as bicubic downsampling, to the available HR images. These fabricated LR images are used as a true LR-HR pair in supervised training which yields poor generalization to real-world LR observations. Additionally, collecting large datasets of authentic LR-HR pairs, particularly for clinical images, is both challenging and costly. We propose an unsupervised transformer-based domain adaptation SR model TUDASR for an arbitrary set of WCE images. The architecture makes use of residual design along with a transformer to extract features associated with local and global features in the LR data. Further, the proposed network is trained on the newly curated dataset of WCE and publicly available conventional endoscopy datasets KVASIR conventional endoscopy dataset to check the efficacy of the proposed method. The visual and quantitative assessments reveal superior SR results using the proposed method over the other existing SR models, for instance, the proposed method works better in non-reference quantitative measures i.e., BRISQUE and PIQE.