Pickled cucumber is a favorite pickle product of the general public, but the long-term or excessive consumption of traditional high-salt pickles seriously threatens people’s health. In this paper, after the subject samples were pickled according to the process flow, one-way and orthogonal tests were designed to experiment with the optimum additive amount of composite pickling agent for cucumber, and meanwhile, the amino acid content of cucumber in different pickling methods was calculated. The Gaussian process regression model’s learning process was optimized by choosing the squared exponential covariance function. The cuckoo search algorithm is proposed, and the improvement of the cuckoo search algorithm is achieved by combining step length adaptive change calculation and chaotic sequences. Jointly improve the cuckoo search algorithm and Gaussian process algorithm, construct the agent model of the pickled cucumber recipe, get the final pickled cucumber recipe parameter design results, and verify the optimization effect of the process through the control experiment. The experimental results show that the best pickling combination is the composite pickling agent with 0.1% addition of acetic acid, 1.5% addition of mannitol and 0.06% addition of calcium lactate, and the sensory score of the cucumber pickled from that place is 92.8564. After optimization, the amino acid content of the cucumber pickled using the composite pickling agent is only 0.03889g/100g, and the quality of the cucumber has been improved.