Aims <p>Carbon cycling is crucial for forest ecosystem function and stability. Litter and root inputs, as the main sources in the soil carbon cycle, play a key role in soil carbon accumulation in arid zones. We aimed at determining the effects of litter and root inputs on the soil microbial carbon cycle in arid areas.</p> Methods <p>Using metagenomic technology, combined with R software, hierarchical partition analysis, network analysis, and correlation analysis, we investigated the results of a six-year plant residue addition and removal experiment and compared the microbial communities, pathways and associated genes, and network patterns related to soil carbon cycling under litter and root inputs.</p> Results <p>No root treatment promoted an increase in the diversity of microorganisms involved in the carbon cycle. Double litter treatment increased the abundance of genes related to carbon fixation and starch degradation. No input treatment significantly reduced the abundance of carbon degradation and carbon fixation genes, while No input treatment increased the abundance of genes involved in hemicellulose degradation, with the abundance of <i>abfA</i> and <i>MAN2C</i> genes increasing by 166% and 481%, respectively, compared to the Control treatment. Among the networks, no root treatment possessed the highest network topological properties. Key genes involved in the carbon cycle were mainly from the carbon fixation pathway, and key genes were mainly affected by electrical conductivity and pH.</p> Conclusions <p>This study provides important information for understanding the carbon cycle process of mountain forest ecosystems in arid areas.</p>