Tool mode runs a fixed set of scanpy functions, so the same request behaves the same way every time; code mode hands the model full Python and covers anything the predefined functions don't, at the cost of varying between runs. If the analysis ends up in a paper, the authors ask that you cite scanpy, LIANA+, decoupleR and CellRank 2.
A Python server that puts a scRNA-seq workflow behind natural language: reading and writing data, preprocessing, clustering, differential expression, plotting, cell-cell communication, pseudotime and enrichment.
- An IO module for reading and writing scRNA-seq data
- Preprocessing: filtering, quality control, normalization, scaling, highly-variable genes, PCA and neighbors
- Analysis: clustering and differential expression, plus cell-cell communication, pseudotime and enrichment
- Plots: violin plots, heatmaps and dotplots
- Two run modes — tool mode calls predefined functions; code mode gives the model a Jupyter backend and returns a complete notebook with executable code, results and figures
`pip install scmcp`, then `scmcp run --run-mode tool` or `scmcp run --run-mode code`. For remote use, `scmcp run --transport shttp --port 8000` and point the client at `http://localhost:8000/mcp`.
One command — pip install scmcp