ncbi-kg-mcp 1.0.1

MCP server for the NCBI Knowledge Graph - query 82K biomedical nodes via any LLM

ncbi-kg-mcp

MCP server for the NCBI Knowledge Graph - a BioLink-compliant glucose metabolism graph with 82,517 nodes and 263,404 edges. Query genes, diseases, variants, proteins, pathways, and more from any LLM.

Install

pip install ncbi-kg-mcp

Configure

Add to your .mcp.json (Claude Code, Cursor, Cline, etc.):

{
  "mcpServers": {
    "ncbi-kg": {
      "command": "ncbi-kg-mcp"
    }
  }
}

That's it. No API keys, no database credentials, no repo cloning needed.

Tools (5)

Tool	Input	Description
query	question: str	Ask a natural language question - auto-translated to Cypher
cypher	query_text: str	Run a read-only Cypher query directly
get_stats	None	Get node and edge counts by type
get_schema	None	Get graph schema - node types, relationships, patterns
get_neighbors	node_id: str	Get 1-hop neighbors of a node (e.g., "NCBIGene:2645")

Example

Ask your LLM: "What genes are associated with MODY?"

The LLM calls the query tool, which translates to Cypher, executes against Neo4j, and returns a synthesized answer with raw data.

Graph Data

• 8 node types: Gene, Disease, SequenceVariant, BiologicalProcess, MolecularActivity, Protein, CellularComponent, Pathway
• 15 relationship types: GENE_ASSOCIATED_WITH_CONDITION, IS_SEQUENCE_VARIANT_OF, CAUSES, INVOLVED_IN, ENABLES, and more
• Sources: NCBI Gene, ClinVar, MedGen, MONDO, GO, UniProt, Reactome

Configuration

Env Variable	Default	Description
NCBI_KG_API_URL	https://ncbi-kg-api-production.up.railway.app	API base URL

Override the API URL to point to a local backend:

{
  "mcpServers": {
    "ncbi-kg": {
      "command": "ncbi-kg-mcp",
      "env": {"NCBI_KG_API_URL": "http://localhost:8001"}
    }
  }
}

Links

• Full documentation
• REST API reference
• Live API docs