stMCP: spatial transcriptomics with a Model Context Protocol server

Spatial transcriptomics enables high-resolution mapping of gene expression within intact tissues but remains computationally challenging, limiting accessibility and iterative analysis for researchers without extensive programming experience. We developed stMCP, a Model Context Protocol (MCP) server that provides a natural language interface for spatial transcriptomics analysis while separating orchestration from locally executed computational workflows. stMCP integrates a language model interface, modular orchestration layer, predefined analytical tools, session management, comprehensive analysis logging, and a web-based results viewer. Analytical operations are performed locally, allowing biological data to remain on the user’s system while reducing repeated language-model interaction during computation. A streamlined installation workflow further reduces the software configuration required for deployment. Benchmarking demonstrated robust orchestration across common spatial transcriptomics tasks, including recovery from ambiguous or unsupported requests. stMCP successfully analyzed larger public Xenium datasets and supported workflows across multiple spatial transcriptomics platforms. Direct comparison with STAgent and SpatialAgent further demonstrated differences in execution efficiency, language-model utilization, reproducibility, and deployment requirements between constrained MCP-based orchestration and autonomous agent-based analysis. stMCP provides an accessible and reproducible interface for spatial transcriptomics analysis while retaining established computational methods and local control of analytical execution. More broadly, the framework demonstrates how MCP-based architecture can connect natural language interfaces with scientific computing tools while balancing accessibility, computational control, and reproducibility.