At the mesoscale, reaction networks are described in terms of stochastic processes. In well-stirred solutions, the time evolution is ruled by the chemical master equation for the probability distribution of the random numbers of molecules. The entropy production is obtained for these reactive processes in the framework of stochastic thermodynamics. The entropy production can be decomposed using the Hill–Schnakenberg cycle decomposition in terms of the affinities and the reaction rates of the stoichiometric cycles of the reaction network. The multivariate fluctuation relation is established for the reactive currents. The results are applied to several examples of reaction networks, in particular, describing autocatalytic bistability, noisy chemical clocks, enzymatic kinetics, and copolymerization processes.