Wrapping preprocessing and plotting alongside the analysis step is what makes this usable end to end — a clustering call is worth little if getting the data normalised still means dropping back to a notebook. The remote transport is the practical detail for this field: single-cell work happens on machines with the memory for it, and this lets the client stay local while the computation does not.
A natural-language interface to CellRank for scRNA-seq work. It spans the analysis pipeline rather than one step: reading and writing data, preprocessing, the analysis tools, and plotting — usable from AI clients, editor plugins and agent frameworks alike.
- An IO module for reading and writing scRNA-seq data
- A preprocessing module covering filtering, quality control, normalization, scaling, highly-variable gene selection, PCA and neighbours
- An analysis module including clustering and differential expression
- A plotting module for violin plots, heatmaps and dotplots
- Two transports: stdio for a local client, or Streamable HTTP when the analysis runs on a server and the client is on your laptop
`pip install cellrank-mcp` from PyPI, then `cellrank-mcp run` to check it works. For a local client, point the config at the installed binary with `run` as its argument — `which cellrank` finds the path. To run it remotely, launch `cellrank-mcp run --transport shttp --port 8000` on the server and give the client the `/mcp` URL. Full documentation is at docs.scmcphub.org. If you use it in research, the README asks you to cite the CellRank 2 paper in Nature Methods.
One command — pip install cellrank-mcp