We present the design and implementation of Mopsa, a platform that simplifies the construction of semantic static analyzers by abstract interpretation. Mopsa computes sound program invariants and reports run-time errors, undefined behaviors, and uncaught exceptions. Mopsa differs from existing platforms by its highly modular and extensible design: semantic abstractions of numeric values, pointers, objects, control, as well as syntax-driven iterators, are defined in small, reusable domains with loose coupling, that can be combined and reused to a greater extent than in previous work. Moreover, Mopsa aims at supporting several languages (currently, subsets of both C and Python) while sharing abstraction components as much as possible. Mopsa is a work in progress, and not yet capable of analyzing full programs; nevertheless, we report early experimental results on verification benchmarks.