Here we present a modified version of the on the-fly string method for the localization of the minimum free energy path in a space of arbitrary collective variables. In the proposed approach the shape of the biasing potential is controlled by only two force constants, defining the width of the potential along the string and orthogonal to it. The force constants and the distribution of the string nodes are optimized during the simulation, improving the convergence. The optimized parameters can be used for umbrella sampling with a path CV along the converged string as the reaction coordinate. We test the new method with three fundamentally different processes: chloride attack to chloromethane in bulk water, alanine dipeptide isomerization, and the enzymatic conversion of isochorismate to piruvate. In each case the same set of parameters resulted in a rapidly converging simulation and a precise estimation of the potential of mean force. Therefore, the default settings can be used for a wide range of processes, making the method essentially parameter free and more user-friendly.