Metacognitive feelings are affective experiences that concern the subject's mental processes and capacities. They include feelings like familiarity, confidence, and the tip-of-the-tongue experience. This article presents a predictive-processing (PP) account of metacognitive feelings, which posits that the brain is a hierarchical prediction machine that predicts sensory input based on prior experience and updates predictions based on prediction errors. Metacognitive feelings arise from visceral changes that serve as cues to predict error dynamics in mental processes. The expected rate of prediction-error reduction corresponds to the valence of the metacognitive feeling. These feelings model the agent's situation both descriptively and directive, serving as appraisals of cognitive performance and action policies. Metacognitive feelings are not only evaluations of cognitive performance but also guide mental and physical actions. They are conscious experiences with valence, often linked to interoceptive inference and affective dynamics. The PP framework suggests that metacognitive feelings emerge from the evaluation of prediction error dynamics, with valence corresponding to the expected rate of error reduction. This framework integrates interoception, prediction error dynamics, and action policies, providing a unified account of metacognitive feelings as conscious, valenced experiences that guide cognitive and behavioral processes. The article also discusses how metacognitive feelings influence mental action, including the selection of action policies based on predicted outcomes and the role of precision weighting in adjusting cognitive processes. The PP perspective highlights the importance of interoceptive inference and the dynamic modeling of prediction errors in understanding metacognitive feelings as conscious, valenced experiences that guide cognitive and behavioral processes.Metacognitive feelings are affective experiences that concern the subject's mental processes and capacities. They include feelings like familiarity, confidence, and the tip-of-the-tongue experience. This article presents a predictive-processing (PP) account of metacognitive feelings, which posits that the brain is a hierarchical prediction machine that predicts sensory input based on prior experience and updates predictions based on prediction errors. Metacognitive feelings arise from visceral changes that serve as cues to predict error dynamics in mental processes. The expected rate of prediction-error reduction corresponds to the valence of the metacognitive feeling. These feelings model the agent's situation both descriptively and directive, serving as appraisals of cognitive performance and action policies. Metacognitive feelings are not only evaluations of cognitive performance but also guide mental and physical actions. They are conscious experiences with valence, often linked to interoceptive inference and affective dynamics. The PP framework suggests that metacognitive feelings emerge from the evaluation of prediction error dynamics, with valence corresponding to the expected rate of error reduction. This framework integrates interoception, prediction error dynamics, and action policies, providing a unified account of metacognitive feelings as conscious, valenced experiences that guide cognitive and behavioral processes. The article also discusses how metacognitive feelings influence mental action, including the selection of action policies based on predicted outcomes and the role of precision weighting in adjusting cognitive processes. The PP perspective highlights the importance of interoceptive inference and the dynamic modeling of prediction errors in understanding metacognitive feelings as conscious, valenced experiences that guide cognitive and behavioral processes.