Short peptides offer a cheap alternative to antibodies for developing sensing units in devices for concentration measurement. We here describe a computational procedure that allows designing peptides capable of binding with high affinity a target organic molecule in aqueous or nonstandard solvent environments. The algorithm is based on a stochastic search in the space of the possible sequences of the peptide, and exploits finite temperature molecular dynamics simulations in explicit solvent to check if a proposed mutation improves the binding affinity or not. The procedure automatically produces peptides which form thermally stable complexes with the target. The estimated binding free energy reaches the 13 kcal/mol for Irinotecan anticancer drug, the target considered in this work. These peptides are by construction solvent specific; namely, they recognize the target only in the solvent in which they have been designed. This feature of the algorithm calls for applications in devices in which the peptide-based sensor is required to work in denaturants or under extreme conditions of pressure and temperature.