Maftools-MCP: A Model Context Protocol–enabled Framework for LLM–driven Somatic Mutation Analysis
Ching-Hsuan Lin, Chi-Ching Lee, Sin-Hong Shih, Cheng-Chi Yang, Mu-Yun Xu, Yuan-Ming Yeh, Po-Jung HuangAbstract
Motivation
Somatic mutation analysis is routine in cancer genomics, but the scale and complexity of next-generation sequencing data can make analysis challenging. Maftools is widely used for somatic variant analysis and visualization, yet its R-based workflow can be difficult for non-programmers, while graphical interfaces may provide access to only predefined analytical options.
Results
We developed Maftools-MCP, a framework based on the Model Context Protocol that translates natural-language requests into structured tool calls for Maftools analysis. Maftools-MCP demonstrates how this protocol can enable natural-language interaction with an R-based analytical package while allowing flexible control of exposed functions and parameters. We also developed an agent skill to support upstream variant annotation before downstream Maftools analysis. Evaluation included per-tool validation, repeated execution of three representative use cases across five language models, including commercial and locally deployed models, and comparison of an integrated annotation workflow with a directly executed reference workflow.
Availability and implementation
Maftools-MCP is publicly available at http://tardis.cgu.edu.tw/maftools-mcp.