The Neo4j MCP Servers connect large language models to Neo4j graph databases through the Model Context Protocol, letting AI assistants run queries, explore schemas, and work with graph data as part of an agent workflow. Developed by Neo4j Labs and the Field GenAI team, the project is MIT licensed and has earned just under 1,000 stars on GitHub, with active updates continuing through 2026.
The repository ships a family of servers rather than a single tool: servers for running Cypher queries against a database, reading graph data, managing memory-style records for agents, and related utilities that expose Neo4j to any MCP-compatible client over stdio. Because they follow the MCP standard, the same servers plug into desktop AI apps, coding agents, and custom agent stacks alike. As a Neo4j Labs project, the code is community-supported and evolves quickly โ the team notes that features are experimental and backwards compatibility is not guaranteed, which is a fair trade for early access to graph-aware agents maintained by the people who build Neo4j itself.
Graph data is a natural fit for LLM agents: questions about relationships, paths, and neighborhoods that are awkward in relational terms become single Cypher queries, and the results come back as structured records an agent can reason over. The memory server in particular points at a growing pattern โ giving agents a persistent, queryable graph of past interactions and facts instead of a flat chat history. Together, the servers make Neo4j a practical context layer for agent applications rather than just a system of record sitting behind them.
The Neo4j MCP Servers are free and open source under the MIT license. You bring your own Neo4j database, which has its own free and commercial editions.
They are aimed at developers building graph-backed AI agents and retrieval systems who want their LLM to query Neo4j directly instead of through hand-written glue code.