OmicVerse MCP Server¶

The OmicVerse MCP server exposes registered analysis functions as tools via the Model Context Protocol. Claude Code, Claude Desktop, and other MCP clients can discover and call these tools while the underlying AnnData objects stay server-side.

How to read the MCP docs now

This page is now the overview entrypoint. The original long guide has been split by topic so you can go directly to the part you need without reading one very large page.

Reading Map¶

Fastest path: Quick Start
Complete onboarding path: Full Start
Full tool inventory: Tool Catalog
Claude / HTTP / SSH setup: Clients and Deployment
adata_id, persistence, cancel, logs, troubleshooting: Runtime and Troubleshooting
Flags, JSON-RPC, error codes, exact counts: Reference
Claude Code walkthrough with screenshots: Using OmicVerse MCP with Claude Code

What OmicVerse MCP Provides¶

OmicVerse analysis tools exposed as MCP tools
Stable handles such as adata_id, artifact_id, and instance_id
Built-in datasets such as ov.datasets.pbmc3k, ov.datasets.pbmc8k, and ov.datasets.seqfish
AnnData inspection tools such as ov.adata.describe, ov.adata.peek, and ov.adata.inspect
Two local transports: stdio and streamable-http

Current Scope¶

With the current implementation:

P0: 15 analysis tools
P0.5: 13 additional analysis tools
P2: 5 advanced class-backed tools
Meta tools: 25 always-on tools
Total at P0+P0.5+P2: 58 tools

What Moved Where¶

Installation, startup, and first analysis¶

Moved to:

Phase system, tool catalog, and dependency tiers¶

Moved to:

Tool Catalog

Client configuration¶

Moved to:

Clients and Deployment

This includes:

stdio
local streamable-http
Claude Code / Claude Desktop
remote SSH launch
local and remote side-by-side deployment

Sessions, handles, persistence, observability, and troubleshooting¶

Moved to:

Runtime and Troubleshooting

JSON-RPC and raw client examples¶

Moved to:

Reference

Minimal Example¶

If you only want the shortest useful flow:

Load the built-in pbmc3k dataset
Describe the current adata
Run QC, scale, PCA with 50 components, build neighbors, compute UMAP, and run Leiden clustering
Plot the UMAP colored by leiden

Use Quick Start for the command and config snippets behind this flow.