What this removes is the scaffolding: loading, calling the method with the right arguments, and getting a circle plot out. For exploring a clustered dataset conversationally, or for an agent that needs communication inference as one step in a longer pipeline, that is exactly the friction worth removing. It is a wrapper, so the method's assumptions and its limits are LIANA+'s — read them there before you draw a conclusion from an edge on a circle plot.
An MCP server wrapping LIANA+, the framework for inferring cell-cell communication from single-cell RNA-seq data. It covers reading and writing the data, running the communication inference, and producing the plots the result is normally read from.
- Single-cell data read in and written back out — the IO module
- Cell-cell communication inference over the loaded object — the method module
- The two plots the results are usually read from: circle plot and dotplot — the plotting module
- Use from most clients that speak MCP, including desktop clients, editor plugins and agent frameworks
No account and no key — it runs against data on your own machine. `pip install liana-mcp`, then `liana-mcp run` starts it over stdio; the client entry points at the absolute path of the installed binary with the `run` argument. To reach it from another machine, `liana-mcp run --transport shttp --port 8000` serves it at `http://localhost:8000/mcp`. The pyproject name is `liana-mcp`. Documentation for the scmcphub family is at docs.scmcphub.org. If you publish work using it, the authors ask that you cite the LIANA+ paper.
One command — pip install liana-mcp