Efficient modeling approaches are necessary to accurately predict large-scale structural behavior of biomolecular systems like RNA (ribonucleic acid). Coarse-grained approximations of such complex systems can significantly reduce the computational costs of the simulation while maintaining sufficient fidelity to capture the biologically significant motions. However, given the coupling and nonlinearity of RNA systems (and effectively all biopolymers), it is expected that different parameters such as geometric and dynamic boundary conditions, and applied forces will affect the system's dynamic behavior. Consequently, static coarse-grained models (i.e., models for which the coarse graining is time invariant) are not always able to adequately sample the conformational space of the molecule. We introduce here the concept of adaptive coarse-grained molecular dynamics of RNA, which automatically adjusts the coarseness of the model, in an effort to more optimally increase simulation speed, while maintaining accuracy. Adaptivity requires two basic algorithmic developments: first, a set of integrators that seamlessly allow transitions between higher and lower fidelity models while preserving the laws of motion. Second, we propose and validate metrics for determining when and where more or less fidelity needs to be integrated into the model to allow sufficiently accurate dynamics simulation. Given the central role that multibody dynamics plays in the proposed framework, and the nominally large number of dynamic degrees of freedom being considered in these applications, a computationally efficient multibody method which lends itself well to adaptivity is essential to the success of this effort. A suite of divide-and-conquer algorithm (DCA)-based approaches is employed to this end. These algorithms have been selected and refined for this purpose because they offer a good combination of computational efficiency and modular structure.