Integrated bioengineering systems can make executable decisions according to the cell state. To sense the state, multiple biomarkers are detected and processed via logic gates with synthetic biological devices. However, numerical operations have not been achieved. Here, we show a design principle for messenger RNA (mRNA) devices that recapitulates intracellular information by multivariate calculations in single living cells. On the basis of this principle and the collected profiles of multiple microRNA activities, we demonstrate that rationally programmed mRNA sets classify living human cells and track their change during differentiation. Our mRNA devices automatically perform multivariate calculation and function as a decision-maker in response to dynamic intracellular changes in living cells.