An MCP server for scRNA-Seq analysis with natural language!

• IO module: Read and write scRNA-Seq data with natural language
• Preprocessing module: Filtering, quality control, normalization, scaling, highly-variable genes, PCA, Neighbors,...
• Tool module: Clustering, differential expression, etc.
• Plotting module: Violin plots, heatmaps, dotplots
• Cell-cell communication analysis
• Pseudotime analysis
• Enrichment analysis

• Anyone who wants to do scRNA-Seq analysis using natural language!
• Agent developers who want to call scanpy's functions for their applications

You can use scmcp in most AI clients, plugins, or agent frameworks that support the MCP:

• AI clients, like Cherry Studio
• Plugins, like Cline
• Agent frameworks, like Agno

scmcphub's complete documentation is available at https://docs.scmcphub.org

A demo showing scRNA-Seq cell cluster analysis in an AI client Cherry Studio using natural language based on scmcp:

Install from PyPI:

pip install scmcp

You can test it by running:

scmcp run

SCMCP provides two distinct run modes to accommodate different user needs and preferences:

In tool mode, SCMCP provides a curated set of predefined functions that the LLM can select and execute.

Advantages:

• Stable: Predefined functions ensure consistent and reliable execution
• Predictable: Known behavior and expected outputs
• Safe: Controlled environment with validated operations

Disadvantages:

• Limited flexibility: Restricted to available predefined functions; you need to define new tools when you need customization functions

Running in terminal:

scmcp run --run-mode tool

Configure MCP client:

{
  "mcpServers": {
    "scmcp": {
      "command": "/home/test/bin/scmcp",
      "args": ["run", "--run-mode", "tool"]
    }
  }
}

Examples:

• https://youtu.be/VM7G-VMNhOs
• https://youtu.be/D669c0EyNzE
• https://youtu.be/b45_6fJXEIQ

In code mode, SCMCP provides a Jupyter backend that allows the LLM to generate and execute custom code. Additionally, it can generate complete Jupyter notebooks containing executable code, analysis results, and visualizations.

This mode is based on the project: https://github.com/huang-sh/abcoder

Advantages:

• Highly flexible: Can create custom workflows and combine operations freely
• Extensible: Supports any Python code and external libraries
• Interactive: Real-time code execution and debugging capabilities

Disadvantages:

• Less stable: Code generation may vary each time

Running in terminal:

scmcp run --run-mode code

Configure MCP client:

{
  "mcpServers": {
    "scmcp": {
      "command": "/home/test/bin/scmcp",
      "args": ["run", "--run-mode", "code"]
    }
  }
}

Example: https://youtu.be/3jtXIeapslI

For both modes, you can also run SCMCP remotely:

Start the server:

# Tool mode
scmcp run --transport shttp --port 8000 --run-mode tool

# Code mode
scmcp run --transport shttp --port 8000 --run-mode code

Configure your MCP client:

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

If you have any questions, welcome to submit an issue, or contact me (hsh-me@outlook.com). Contributions to the code are also welcome!

If you use scmcp in your research, please consider citing the following works:

Wolf, F., Angerer, P. & Theis, F. SCANPY: large-scale single-cell gene expression data analysis. Genome Biol 19, 15 (2018). https://doi.org/10.1186/s13059-017-1382-0

Dimitrov D., Schäfer P.S.L, Farr E., Rodriguez Mier P., Lobentanzer S., Badia-i-Mompel P., Dugourd A., Tanevski J., Ramirez Flores R.O. and Saez-Rodriguez J. LIANA+ provides an all-in-one framework for cell–cell communication inference. Nat Cell Biol (2024). https://doi.org/10.1038/s41556-024-01469-w

Badia-i-Mompel P., Vélez Santiago J., Braunger J., Geiss C., Dimitrov D., Müller-Dott S., Taus P., Dugourd A., Holland C.H., Ramirez Flores R.O. and Saez-Rodriguez J. 2022. decoupleR: ensemble of computational methods to infer biological activities from omics data. Bioinformatics Advances. https://doi.org/10.1093/bioadv/vbac016

Weiler, P., Lange, M., Klein, M. et al. CellRank 2: unified fate mapping in multiview single-cell data. Nat Methods 21, 1196–1205 (2024). https://doi.org/10.1038/s41592-024-02303-9