Neo4j Knowledge Graph Memory provides persistent memory infrastructure for AI assistants using Neo4j graph database with semantic search capabilities. It enables AI agents to store, retrieve, and manage interconnected memories across multiple sessions. Key features: - Graph-based memory storage with nodes, relationships, and observations as structured knowledge - Unified semantic search combining vector embeddings (384 dimensions), exact matching, wildcards, and graph traversal - Multi-database project isolation for separating different contexts or users - Smart relationship management with typed connections, strength tracking, and temporal metadata - Batch operations for creating multiple memories with relationships in single requests - Time-based queries using relative ("7d", "30d") or absolute date filters - Graph traversal with directional navigation and depth control - Temporal tracking to monitor how knowledge evolves over time Built for MCP-native integration with Claude Desktop and compatible clients. Requires Neo4j with GDS plugin (recommends DozerDB) for vector operations.

Neo4j Memory Server

A Model Context Protocol (MCP) server that provides AI assistants with persistent, intelligent memory capabilities using Neo4j's graph database with unified architecture

What it does

This server enables AI assistants to:

Features

Core Capabilities

Advanced Operations

Architecture

Technical Highlights

Quick Start

npm install @sylweriusz/mcp-neo4j-memory-server

Add to Claude Desktop config:

{
  "mcpServers": {
    "memory": {
      "command": "npx",
      "args": ["-y", "@sylweriusz/mcp-neo4j-memory-server"],
      "env": {
        "NEO4J_URI": "bolt://localhost:7687",
        "NEO4J_USERNAME": "neo4j", 
        "NEO4J_PASSWORD": "your-password"
      }
    }
  }
}

Neo4j Setup

Working setup: DozerDB with GDS Plugin

For the database, use DozerDB with the Graph Data Science plug-in, GDS is not only recommended but necessary:

For current installation instructions, see: https://dozerdb.org/

Example setup:

# Run DozerDB container with latest version
docker run \
    -p 7474:7474 -p 7687:7687 \
    -v $HOME/neo4j/data:/data \
    -v $HOME/neo4j/logs:/logs \
    -v $HOME/neo4j/plugins:/plugins \
    --env NEO4J_AUTH=neo4j/password \
    --env NEO4J_dbms_security_procedures_unrestricted='gds.*' \
    graphstack/dozerdb:latest
Install GDS plugin - see dozerdb.org for current instructions
Verify GDS plugin works
In Neo4j Browser (http://localhost:7474):
RETURN gds.similarity.cosine([1,2,3], [2,3,4]) as similarity

Unified Tools

The server provides 4 unified MCP tools that integrate automatically with Claude:

Memory Structure

{
  "id": "dZ$abc123",
  "name": "Project Alpha", 
  "memoryType": "project",
  "metadata": {"status": "active", "priority": "high"},
  "observations": [
    {"id": "dZ$obs456", "content": "Started development", "createdAt": "2025-06-08T10:00:00Z"}
  ],
  "related": {
    "ancestors": [{"id": "dZ$def789", "name": "Initiative", "relation": "PART_OF", "distance": 1}],
    "descendants": [{"id": "dZ$ghi012", "name": "Task", "relation": "INCLUDES", "distance": 1}]
  }
}

System Prompt

The simplest use of the memory tool, the following usually is more than enough.

## Memory Tool Usage
Store all memory for this project in database: 'project-database-name'
Use MCP memory tools exclusively for storing project-related information
Begin each session by:
  1. Switching to this project's database
  2. Searching memory for data relevant to the user's prompt

Troubleshooting

Vector Search Issues:

Connection Issues: