Scanpy-MCP

MCP server that exposes the full Scanpy / AnnData workflow as natural-language tools — IO, QC, normalization, PCA, clustering, differential expression, and plotting.

   
Type MCP server
Supplier scmcphub (huang-sh)
Availability GA — v0.5.0 (10 releases)
Pricing Free / OSS (BSD-3-Clause)
Capabilities Read/Write

Install the server:

pip install scanpy-mcp

The scanpy-mcp package exposes two transports:

  • scanpy-mcp run — stdio
  • scanpy-mcp run --transport shttp --port 8000 — streamable HTTP

Register with Claude Code (stdio — recommended; Claude Code launches the process itself, so don’t run scanpy-mcp run separately):

claude mcp add --transport stdio scanpy -- scanpy-mcp run

Or with Claude Desktop, add to claude_desktop_config.json:

{
  "mcpServers": {
    "scanpy": { "command": "scanpy-mcp", "args": ["run"] }
  }
}

Compatible with Claude Desktop, Claude Code, Cherry Studio, Cline, and Agno.

Alternative (skill, not MCP): K-Dense also ships a scanpy Claude Skill that drives Scanpy through local Python instead of an MCP server — npx skills add K-Dense-AI/scientific-agent-skills, then enable the scanpy skill. Same underlying library; pick whichever workflow you prefer.

What it does

  • io.read / io.write — AnnData / 10x file I/O
  • pp.* — QC filter, normalize, find HVGs, scale, PCA, neighbors
  • tl.* — UMAP, Leiden clustering, rank_genes_groups
  • pl.* — violin, heatmap, dotplot

Primary use cases: Natural-language single-cell RNA-seq QC and clustering, reproducible Scanpy pipelines driven from chat, onboarding for users new to Scanpy.

Notes

stdio and streamable-HTTP transports. Complements the single-cell-rna-qc skill: that skill is a workflow checklist that Claude follows; this server lets Claude call Scanpy functions as discrete tools mid-conversation.

Sources

Installed this tool?

Share feedback — install path, OS, errors, workarounds. The form opens with this tool pre-selected and a link back to this page.