Due to the nonsmoothness of the small-signal stability constraint, calculating the available transfer capability (ATC) limited by small-signal stability rigorously through the nonlinear programming is quite difficult. To tackle this challenge, this paper proposes a sequential quadratic programming (SQP) method combined with gradient sampling (GS) in a dual formulation. The highlighted feature is the sample size of the gradient changes dynamically in every iteration, yielding an adaptive gradient sampling (AGS) process. Thus, the computing efficiency is greatly improved owing to the decrease and the parallelization of gradient evaluation, which dominates the computing time of the whole algorithm. Simulations on an IEEE 10-machine 39-bus system and an IEEE 54-machine 118-bus system prove the effectiveness and high efficiency of the proposed method.