The atmospheric component of the Model for Prediction Across Scales (MPAS-A) is noted for an unstructured mesh, but further development is needed to implement more compatible data assimilation techniques. As the Gridpoint Statistical Interpolation (GSI) system that being operational at NCEP features advanced variational data assimilation schemes, a global analysis and forecast framework is constructed to incorporate GSI analysis into MPAS-A in this study. Conservative remapping algorithms are implemented in the framework to handle grid structure transformation between the two components. The performance of the framework is verified by conducting grid transformation tests and cycling experiments. Grid transformation tests show that the remapping errors of model variables are highly related with their distribution patterns. Second-order conservative remapping yields smaller errors than the first-order approach. Continuous cycling experiments are conducted for a week, leading to conclude that the framework could effectively ingest observations of multi sources. With enhanced initial fields, the forecasts of MPAS-A are more accurate, and the modeled precipitation is closer to observations. Further inspections show that improvements of the North hemisphere outperform the South hemisphere and the Tropics, with benefits of more assimilated observations.