Background and purpose To implement a robust and fast stereotactic MR-guided adaptive radiation therapy (SMART) online strategy in locally advanced pancreatic cancer (LAPC). Material and methods SMART strategy for plan adaptation was implemented with the MRIdian system (ViewRay Inc.). At each fraction, OAR (re-)contouring is done within a distance of 3 cm from the PTV surface. Online plan re-optimization is based on robust prediction of OAR dose and optimization objectives, obtained by building an artificial neural network (ANN). Proposed limited re-contouring strategy for plan adaptation (SMART3CM) is evaluated by comparing 50 previously delivered fractions against a standard (re-)planning method using full-scale OAR (re-)contouring (FULLOAR). Plan quality was assessed using PTV coverage (V95%, Dmean, D1cc) and institutional OAR constraints (e.g. V33Gy). Results SMART3CMrequired a significant lower number of optimizations than FULLOAR (4 vs 18 on average) to generate a plan meeting all objectives and institutional OAR constraints. PTV coverage with both strategies was identical (mean V95%= 89%). Adaptive plans with SMART3CMexhibited significant lower intermediate and high doses to all OARs than FULLOAR, which also failed in 36% of the cases to adhere to the V33Gydose constraint. Conclusions SMART3CMapproach for LAPC allows good OAR sparing and adequate target coverage while requiring only limited online (re-)contouring from clinicians.