scVI-Tools MCP

An MCP (Model Context Protocol) server that gives LLMs structured access to scvi-tools knowledge: model documentation, tutorials, API reference, workflow templates, pretrained Hugging Face Hub models, and community FAQ.

No runtime model execution — pure knowledge layer. Works with Claude Desktop, Cursor, and any MCP-compatible client.

Docs: scvi-tools-mcp.readthedocs.io

Quick Start

Claude Desktop

Add to ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "scvi-tools": {
      "command": "uvx",
      "args": ["scvi-tools-mcp"]
    }
  }
}

Claude Code (CLI)

claude mcp add scvi-tools-mcp -s user -- uvx scvi-tools-mcp

OpenAI Codex CLI

Add to ~/.codex/config.toml:

[mcp_servers.scvi-tools-mcp]
command = "uvx"
args   = ["scvi-tools-mcp"]

Cursor / Windsurf / other clients

{ "command": "uvx", "args": ["scvi-tools-mcp"] }

Local install

pip install scvi-tools-mcp
scvi-tools-mcp

Tools

Knowledge Sources

All knowledge is baked into the package as Markdown files at build time. No network calls at tool-call time.

Knowledge Refresh (CI)

Automated GitHub Actions jobs keep knowledge current — each opens a PR if a diff is found:

All workflows also support workflow_dispatch for manual runs.

Development

git clone https://github.com/Yoseflab/scvi-tools-mcp
cd scvi-tools-mcp
pip install -e . --group dev
pytest

Rebuild knowledge manually

# Convert tutorials from a local scvi-tools checkout
python scripts/convert_notebooks.py /path/to/scvi-tools/docs/tutorials/notebooks \
    src/scvi_tools_mcp/knowledge/tutorials

# Extract API docs (requires scvi-tools installed)
pip install -e ".[scvi]"
python scripts/extract_api_docs.py

# Re-scrape external knowledge
python scripts/scrape_external.py

# Refresh Hugging Face Hub model registry snapshot
python scripts/scrape_huggingface_hub.py

Adding a new model

1. Run scripts/extract_api_docs.py after updating scvi-tools.
2. Add the model name to MODEL_NAMES in src/scvi_tools_mcp/tools/_constants.py.
3. Add requirements to MODEL_REQUIREMENTS in _data_prep.py if needed.

License

BSD 3-Clause License

Copyright (c) 2026, Yosef Lab, Weizmann Institute of Science