This study aimed to develop and validate the Metacognitions about Generative AI Use Scale (MGAUS) to assess college students' metacognitive beliefs about generative AI and to explore these metacognitions as predictors of generative AI addiction risk. A total of 1229 college students from China participated in the study, providing data through an online questionnaire. Exploratory factor analysis initially determined the MGAUS's structure, revealing two primary factors: "Positive metacognitions about generative AI use" and "Negative metacognitions about generative AI use", comprising nine items in total. Confirmatory factor analysis further validated the scale's stability and fit, as well as tested measurement invariance across gender, age, and educational levels. Correlation analysis indicated significant positive correlations between both positive and negative metacognitions and generative AI addiction. Additionally, negative metacognitions were significantly positively correlated with anxiety, whereas the correlation between positive metacognitions and anxiety was not significant. Multivariate regression analysis revealed that, after controlling for gender, both positive and negative metacognitions remained significant predictors of generative AI addiction, with negative metacognitions demonstrating stronger predictive power. A network analysis of the scale items further illuminated the close relationship between positive and negative metacognitions. Taken together, these findings contribute to the theoretical understanding of metacognition in the context of generative AI use and provide a scientific foundation for the prevention and intervention of generative AI addiction.