This paper presents a scalable Front Tracking code for simulation of compressible multiphase flows in 3D that uses an octree Adaptive Mesh Refinement (AMR) grid. A Front Tracking code involves many communication routines between and among Lagrangian and Eulerian grids. The parallel performance of a Front Tracking code depends directly on the scalability of these communication routines. Parallelization of operations in AMR Eulerian grid is a well-studied subject and is out of this paper's scope. This paper focuses on achieving scalability of communications involving Lagrangian grid which includes communication between Lagrangian and Eulerian grids in a distributed architecture. In parallel AMR simulation, the Eulerian grid is distributed based on the Z-curve, resulting in a complex domain decomposition which is modified during mesh refinement. The dynamic nature of the grid distribution makes it difficult to synchronize communications between Lagrangian grid and Eulerian grid. This work utilizes distributed fronts (Lagrangian sub-grids) where Eulerian grid cells store Lagrangian grid entities (points and elements), which gives the capability to maintain a direct mapping between Lagrangian points and Eulerian grid cells. Here, mapping allows scaling inter-grid communication routines and reduces the complexity involved in load balancing during refinement. Utilizing the Front Tracking code mentioned above we implement an all-Mach compressible multiphase solver to simulate cavitation of microbubbles. The all-Mach code is written in basilisk \cite{basilisk} takes into account the compressiblity of liquid and surface tension.