PrimeKG‑to‑Neo4j MCP Server

PrimeKG‑to‑Neo4j is an MCP server that bridges the rich, multidimensional biomedical knowledge contained in PrimeKG with the graph‑oriented query capabilities of Neo4j. By exposing PrimeKG data as a persistent, queryable graph through MCP, the server enables AI assistants to retrieve, analyze, and reason over complex biomedical relationships in real time. This solves the common problem of “knowledge silos” where clinical, genomic, and drug‑response data are scattered across disparate files or APIs—making it difficult for AI agents to provide coherent, evidence‑based answers.

The server performs three core functions. First, it automates the download and ETL (extract‑transform‑load) of PrimeKG’s 4 + million relationships and 17,000 disease entities. Second, it materializes the processed data into a Neo4j database, preserving the graph structure and supporting ACID guarantees. Third, it exposes MCP endpoints that allow AI assistants to run Cypher queries, retrieve subgraphs, and perform analytical operations such as shortest‑path discovery or community detection. By exposing these capabilities through MCP, developers can embed sophisticated biomedical reasoning directly into their conversational agents without managing database infrastructure.

Key features include:

• Automated ETL pipeline that keeps the graph up‑to‑date with the latest PrimeKG releases.
• Rich Cypher query interface via MCP, enabling dynamic exploration of disease–gene–drug relationships.
• Pre‑defined analytical tools (e.g., co‑occurrence scoring, pathway enrichment) that can be invoked as MCP tools.
• Scalable Neo4j deployment that supports large‑scale graph analytics and real‑time inference.
• Secure, role‑based access managed through MCP authentication, ensuring sensitive biomedical data remains protected.

Typical use cases span research, clinical decision support, and drug discovery. For example, a clinician assistant can ask an AI: “Show me all drugs that target genes implicated in glioblastoma.” The MCP server translates this into a Cypher query, returns the subgraph, and the AI can synthesize the answer. In drug repurposing pipelines, an assistant might request “Identify potential off‑label therapies for patients with rare metabolic disorders.” The server can perform multi‑hop traversals across disease, gene, and drug nodes to surface candidates. Moreover, the server’s analytical tools allow AI agents to rank candidates based on evidence strength or pathway relevance.

Integrating PrimeKG‑to‑Neo4j into AI workflows is straightforward: developers add the MCP server as a tool source in their assistant configuration, then invoke its endpoints using natural language prompts. The server’s clear, declarative API lets assistants perform complex graph queries without needing to write Cypher themselves. This abstraction lowers the barrier for biomedical developers, accelerates hypothesis generation, and ultimately supports more informed decision making in healthcare settings.