Exploration of obstacle-ridden underwater regions for various marine applications like automated inspection, maintenance and repair of sub-sea structures and search and rescue during disaster relief is often not possible to be carried out by the human divers. Owing to their slender and hyper-redundant structure, Anguilliform-inspired robots are capable of negotiating narrow regions. However, the challenges involved in the motion planning of Anguilliform-inspired robots include the dynamic constraints imposed by the hyper-redundant joints, the interaction between fluid environment and the robot, and the presence of obstacles. This paper reports a model-predictive motion planning approach for an Anguilliform-inspired robot, wherein dynamically feasible motion primitives are generated using a dynamics simulator. The motion primitives are then used for generating a roadmap over which A* algorithm is used for searching an optimal, obstacle-free, and dynamically feasible path to the goal. Use of Euclidean heuristic in the A* based path planning for hyper-redundant underwater robots often results in the expansion of a large number of nodes and thereby slow-down the computations. Hence, we present a simulation-based admissible heuristic function that led to a speed-up of path search computation time by a factor varying from 3.1 to 5.5 over the Euclidean heuristic for our simulation-based experiments. The factor is dependent on the complexity of the scene. We also use dynamics simulation for estimating action-specific convex collision envelops for precise and efficient collision detection during the expansion of nodes in A*.