cellrank-MCP

cellrank-MCP provides a Python-based MCP server that exposes a natural language interface for scRNA-Seq analysis with CellRank. You can run it locally or connect to a remote MCP endpoint to perform IO, preprocessing, analysis, and plotting tasks through simple natural language queries.

How to use

Use an MCP client to talk to the cellrank-MCP server. You can run the server locally for development or deploy it remotely and connect your client via HTTP. Your client can send natural language commands to perform actions such as loading scRNA-Seq data, filtering and normalizing, running clustering or differential expression analyses, and visualizing results.

How to install

Prerequisites: Python and pip must be installed on your system.

pip install cellrank-mcp

Run locally and connect via an MCP client

Start the local stdio MCP server using the explicit runtime path shown in your environment. This config runs the server as a local process that the MCP client communicates with via stdio.

{
  "mcpServers": {
    "cellrank_mcp": {
      "command": "/home/test/bin/cellrank-mcp",
      "args": ["run"]
    }
  }
}

Run remotely and connect via HTTP

Start the remote MCP server and expose its HTTP endpoint for your MCP client to connect.

{
  "mcpServers": {
    "cellrank_mcp": {
      "url": "http://localhost:8000/mcp"
    }
  }
}

Test the server basics

For a quick test, run the server locally first, then configure your MCP client with the appropriate URL. After that, you can issue natural language requests like: load a scRNA-Seq dataset, filter cells, run clustering, and plot violin or heatmap visualizations.

Available tools

IO

Read and write scRNA-Seq data through the server, enabling input/output operations for datasets.

Preprocessing

Quality control, filtering, normalization, scaling, feature selection, and dimensionality reduction steps like PCA.

Tool

Clustering, differential expression, and other analysis steps accessible via natural language commands.

Plotting

Visualization tools such as violin plots, heatmaps, dot plots, and related figures.