The introduction of the MR-linac marked a major milestone in the field of radiotherapy. The MR-linac uses a linear accelerator (Linac) combined with an MRI scanner. MRI allows excellent soft-tissue visualization, and the integrated system enables 3D imaging immediately before and during treatment. Finally we can distinguish tumor from the healthy tissues while the patient is on the table and thus irradiate the tumor much more precisely and safely. Based on these images, the treatment plan can be adapted to the patient’s anatomy at the time of treatment while the patient is still on the treatment table, or even while the radiation is being delivered. This process is known as adaptive radiotherapy. Adaptive radiotherapy treatments require Quality Assurance (QA) to verify that the radiation dose is delivered to the correct location and with the correct quantity. Every step of the online adaptive workflow must be validated to maintain the safety and effectiveness of the treatment. This thesis focuses on validating the online adaptive workflow of the MR-linac. First, system-related uncertainties in dose calculations were investigated. Next, the online adaptive workflow itself was assessed. Finally, the development of phantoms and detectors to enable QA for treatments involving moving, deformable tumors was explored. Several key conclusions were drawn from the studies conducted. First it was shown that dosimetric uncertainties are small and consistent, regardless of tumor location. Next, a total treatment with multiple adaptive fractions resulted in comparable dosimetric uncertainties to separate non-adaptive fractions. Furthermore, when using accumulated dose distributions to evaluate treatments, spatial uncertainties arise, which can be quantified using the newly-developed δ index. It was also demonstrated that intra-fraction plan adaptations during prostate treatments lead to better dosimetric outcomes compared to treatments without plan adaptations. Finally, we validated the initial performance of a novel MRI-compatible prototype of a deformable motion phantom with integrated plastic scintillation dosimeters. This phantom could make an important contribution to the validation and subsequent clinical introduction of new techniques. Additionally, we showed that these dosimeters can be efficiently calibrated with a novel, faster method without loss of dose accuracy. Overall, this thesis has quantified the uncertainties associated with dose delivery in online adaptive treatments. The workflows and adaptation methods on the MR-linac will continue to evolve. As these innovations become standard practice, it is essential for QA protocols to evolve in parallel, meeting the demands of modern radiotherapy.