The present article is devoted to develop an adaptive scheme for the numerical solution of fractional integro-differential equations with weakly singular kernel. The adaptive scheme is based on the product integration method of Huber. An error estimate is provided for the discretisation error occurring at each step to calculate the step size of next integration step. A parameter known as error tolerance is predefined to control local discretisation error. Computations verify that by controlling local discretisation error, the true global errors match fairly well to the error tolerance parameter. The first integration step size $$h_{{ initial}}$$ is introduced in order to make a controlled evaluation of local discretisation error at the first integration step also, where the error estimate is not available. Finally, the computations and results of some numerical experiments validate the accuracy and applicability of the adaptive scheme.