Agent Frameworks

Agent frameworks give developers the building blocks to go beyond single LLM calls — connecting language models to tools, memory, and external data sources. Instead of a linear prompt-response cycle, agents can reason over multiple steps, choose which tools to invoke, and adapt their behaviour based on intermediate results.

Neo4j fits naturally into agent workflows as both a knowledge source and a memory store. Agents query the graph through MCP tools or direct driver integrations to retrieve structured context, traverse relationships across entities, and ground responses in real data rather than model weights alone.

The connection point is usually the Neo4j MCP server, which exposes schema inspection and Cypher execution as standard tools that any MCP-compatible agent can call. For frameworks that prefer direct integration, the Neo4j Python driver and langchain-neo4j / llama-index-graph-stores-neo4j packages provide native tool definitions and vector store support.