icon for mcp server

Neo4j Knowledge Graph Memory

STDIO

Neo4j graph database MCP server providing persistent memory capabilities with relationship tracking and knowledge storage

🧠🕸️ Neo4j Knowledge Graph Memory MCP Server

🌟 Overview

A Model Context Protocol (MCP) server implementation that provides persistent memory capabilities through Neo4j graph database integration.

By storing information in a graph structure, this server maintains complex relationships between entities as memory nodes and enables long-term retention of knowledge that can be queried and analyzed across multiple conversations or sessions.

With Neo4j Aura you can host your own database server for free or share it with your collaborators. Otherwise you can run your own Neo4j server locally.

The MCP server leverages Neo4j's graph database capabilities to create an interconnected knowledge base that serves as an external memory system. Through Cypher queries, it allows exploration and retrieval of stored information, relationship analysis between different data points, and generation of insights from the accumulated knowledge. This memory can be further enhanced with Claude's capabilities.

🕸️ Graph Schema

  • Memory - A node representing an entity with a name, type, and observations.
  • Relationship - A relationship between two entities with a type.

🔍 Usage Example

Let's add some memories 
I, Michael, living in Dresden, Germany work at Neo4j which is headquartered in Sweden with my colleagues Andreas (Cambridge, UK) and Oskar (Gothenburg, Sweden)
I work in Product Management, Oskar in Engineering and Andreas in Developer Relations.

Results in Claude calling the create_entities and create_relations tools.

📦 Components

🔧 Tools

The server offers these core tools:

🔎 Query Tools

  • read_graph

    • Read the entire knowledge graph
    • No input required
    • Returns: Complete graph with entities and relations

  • search_nodes

    • Search for nodes based on a query
    • Input:
      • query (string): Search query matching names, types, observations
    • Returns: Matching subgraph

  • find_nodes

    • Find specific nodes by name
    • Input:
      • names (array of strings): Entity names to retrieve
    • Returns: Subgraph with specified nodes

♟️ Entity Management Tools

  • create_entities

    • Create multiple new entities in the knowledge graph
    • Input:
      • entities: Array of objects with:
        • name (string): Name of the entity
        • type (string): Type of the entity
        • observations (array of strings): Initial observations about the entity
    • Returns: Created entities

  • delete_entities

    • Delete multiple entities and their associated relations
    • Input:
      • entityNames (array of strings): Names of entities to delete
    • Returns: Success confirmation

🔗 Relation Management Tools

  • create_relations

    • Create multiple new relations between entities
    • Input:
      • relations: Array of objects with:
        • source (string): Name of source entity
        • target (string): Name of target entity
        • relationType (string): Type of relation
    • Returns: Created relations

  • delete_relations

    • Delete multiple relations from the graph
    • Input:
      • relations: Array of objects with same schema as create_relations
    • Returns: Success confirmation

📝 Observation Management Tools

  • add_observations

    • Add new observations to existing entities
    • Input:
      • observations: Array of objects with:
        • entityName (string): Entity to add to
        • contents (array of strings): Observations to add
    • Returns: Added observation details

  • delete_observations

    • Delete specific observations from entities
    • Input:
      • deletions: Array of objects with:
        • entityName (string): Entity to delete from
        • observations (array of strings): Observations to remove
    • Returns: Success confirmation

🔧 Usage with Claude Desktop

💾 Installation

pip install mcp-neo4j-memory

⚙️ Configuration

Add the server to your claude_desktop_config.json with configuration of:

"mcpServers": {
  "neo4j": {
    "command": "uvx",
    "args": [
      "[email protected]",
      "--db-url",
      "neo4j+s://xxxx.databases.neo4j.io",
      "--username",
      "<your-username>",
      "--password",
      "<your-password>"
    ]
  }
}

Alternatively, you can set environment variables:

"mcpServers": {
  "neo4j": {
    "command": "uvx",
    "args": [ "[email protected]" ],
    "env": {
      "NEO4J_URL": "neo4j+s://xxxx.databases.neo4j.io",
      "NEO4J_USERNAME": "<your-username>",
      "NEO4J_PASSWORD": "<your-password>"
    }
  }
}

Namespacing

For multi-tenant deployments, add --namespace to prefix tool names:

"args": [ "[email protected]", "--namespace", "myapp", "--db-url", "..." ]

Tools become: myapp-read_graph, myapp-create_entities, etc.

Can also use NEO4J_NAMESPACE environment variable.

🌐 HTTP Transport Mode

The server supports HTTP transport for web-based deployments and microservices:

# Basic HTTP mode (defaults: host=127.0.0.1, port=8000, path=/mcp/)
mcp-neo4j-memory --transport http

# Custom HTTP configuration
mcp-neo4j-memory --transport http --host 127.0.0.1 --port 8080 --path /api/mcp/

Environment variables for HTTP configuration:

export NEO4J_TRANSPORT=http
export NEO4J_MCP_SERVER_HOST=127.0.0.1
export NEO4J_MCP_SERVER_PORT=8080
export NEO4J_MCP_SERVER_PATH=/api/mcp/
export NEO4J_NAMESPACE=myapp
mcp-neo4j-memory

🔄 Transport Modes

The server supports three transport modes:

  • STDIO (default): Standard input/output for local tools and Claude Desktop
  • SSE: Server-Sent Events for web-based deployments
  • HTTP: Streamable HTTP for modern web deployments and microservices

🐳 Using with Docker

"mcpServers": {
  "neo4j": {
    "command": "docker",
    "args": [
      "run",
      "--rm",
      "-e", "NEO4J_URL=neo4j+s://xxxx.databases.neo4j.io",
      "-e", "NEO4J_USERNAME=<your-username>",
      "-e", "NEO4J_PASSWORD=<your-password>",
      "mcp/neo4j-memory:0.4.1"
    ]
  }
}

🔒 Security Protection

The server includes comprehensive security protection with secure defaults that protect against common web-based attacks while preserving full MCP functionality when using HTTP transport.

🛡️ DNS Rebinding Protection

TrustedHost Middleware validates Host headers to prevent DNS rebinding attacks:

Secure by Default:

  • Only localhost and 127.0.0.1 hosts are allowed by default

Environment Variable:

export NEO4J_MCP_SERVER_ALLOWED_HOSTS="example.com,www.example.com"

🌐 CORS Protection

Cross-Origin Resource Sharing (CORS) protection blocks browser-based requests by default:

Environment Variable:

export NEO4J_MCP_SERVER_ALLOW_ORIGINS="https://example.com,https://app.example.com"

🔧 Complete Security Configuration

Development Setup:

mcp-neo4j-memory --transport http \
  --allowed-hosts "localhost,127.0.0.1" \
  --allow-origins "http://localhost:3000"

Production Setup:

mcp-neo4j-memory --transport http \
  --allowed-hosts "example.com,www.example.com" \
  --allow-origins "https://example.com,https://app.example.com"

🚨 Security Best Practices

For allow_origins:

  • Be specific: ["https://example.com", "https://example.com"]
  • Never use "*" in production with credentials
  • Use HTTPS origins in production

For allowed_hosts:

  • Include your actual domain: ["example.com", "www.example.com"]
  • Include localhost only for development
  • Never use "*" unless you understand the risks

🐳 Docker Deployment

The Neo4j Memory MCP server can be deployed using Docker for remote deployments. Docker deployment should use HTTP transport for web accessibility. In order to integrate this deployment with applications like Claude Desktop, you will have to use a proxy in your MCP configuration such as mcp-remote.

📦 Using Your Built Image

After building locally with docker build -t mcp-neo4j-memory:latest .:

# Run with http transport (default for Docker)
docker run --rm -p 8000:8000 \
  -e NEO4J_URI="bolt://host.docker.internal:7687" \
  -e NEO4J_USERNAME="neo4j" \
  -e NEO4J_PASSWORD="password" \
  -e NEO4J_DATABASE="neo4j" \
  -e NEO4J_TRANSPORT="http" \
  -e NEO4J_MCP_SERVER_HOST="0.0.0.0" \
  -e NEO4J_MCP_SERVER_PORT="8000" \
  -e NEO4J_MCP_SERVER_PATH="/mcp/" \
  mcp/neo4j-memory:latest

# Run with security middleware for production
docker run --rm -p 8000:8000 \
  -e NEO4J_URI="bolt://host.docker.internal:7687" \
  -e NEO4J_USERNAME="neo4j" \
  -e NEO4J_PASSWORD="password" \
  -e NEO4J_DATABASE="neo4j" \
  -e NEO4J_TRANSPORT="http" \
  -e NEO4J_MCP_SERVER_HOST="0.0.0.0" \
  -e NEO4J_MCP_SERVER_PORT="8000" \
  -e NEO4J_MCP_SERVER_PATH="/mcp/" \
  -e NEO4J_MCP_SERVER_ALLOWED_HOSTS="example.com,www.example.com" \
  -e NEO4J_MCP_SERVER_ALLOW_ORIGINS="https://example.com" \
  mcp/neo4j-memory:latest

🔧 Environment Variables

VariableDefaultDescription
NEO4J_URIbolt://localhost:7687Neo4j connection URI
NEO4J_USERNAMEneo4jNeo4j username
NEO4J_PASSWORDpasswordNeo4j password
NEO4J_DATABASEneo4jNeo4j database name
NEO4J_TRANSPORTstdio (local), http (remote)Transport protocol (stdio, http, or sse)
NEO4J_MCP_SERVER_HOST127.0.0.1 (local)Host to bind to
NEO4J_MCP_SERVER_PORT8000Port for HTTP/SSE transport
NEO4J_MCP_SERVER_PATH/mcp/Path for accessing MCP server
NEO4J_MCP_SERVER_ALLOW_ORIGINS(empty - secure by default)Comma-separated list of allowed CORS origins
NEO4J_MCP_SERVER_ALLOWED_HOSTSlocalhost,127.0.0.1Comma-separated list of allowed hosts (DNS rebinding protection)
NEO4J_NAMESPACE(empty - no prefix)Namespace prefix for tool names (e.g., myapp-read_graph)

🌐 SSE Transport for Legacy Web Access

When using SSE transport (for legacy web clients), the server exposes an HTTP endpoint:

# Start the server with SSE transport
docker run -d -p 8000:8000 \
  -e NEO4J_URI="neo4j+s://demo.neo4jlabs.com" \
  -e NEO4J_USERNAME="recommendations" \
  -e NEO4J_PASSWORD="recommendations" \
  -e NEO4J_DATABASE="neo4j" \
  -e NEO4J_TRANSPORT="sse" \
  -e NEO4J_MCP_SERVER_HOST="0.0.0.0" \
  -e NEO4J_MCP_SERVER_PORT="8000" \
  --name neo4j-memory-mcp-server \
  mcp-neo4j-memory:latest

# Test the SSE endpoint
curl http://localhost:8000/sse

# Use with MCP Inspector
npx @modelcontextprotocol/inspector http://localhost:8000/sse

🚀 Development

📦 Prerequisites

  1. Install uv (Universal Virtualenv):
# Using pip
pip install uv

# Using Homebrew on macOS
brew install uv

# Using cargo (Rust package manager)
cargo install uv
  1. Clone the repository and set up development environment:
# Clone the repository
git clone https://github.com/yourusername/mcp-neo4j-memory.git
cd mcp-neo4j-memory

# Create and activate virtual environment using uv
uv venv
source .venv/bin/activate  # On Unix/macOS
.venv\Scripts\activate     # On Windows

# Install dependencies including dev dependencies
uv pip install -e ".[dev]"

🐳 Docker

Build and run the Docker container:

# Build the image
docker build -t mcp/neo4j-memory:latest .

# Run the container
docker run -e NEO4J_URL="neo4j+s://xxxx.databases.neo4j.io" \
          -e NEO4J_USERNAME="your-username" \
          -e NEO4J_PASSWORD="your-password" \
          mcp/neo4j-memory:latest

📄 License

This MCP server is licensed under the MIT License. This means you are free to use, modify, and distribute the software, subject to the terms and conditions of the MIT License. For more details, please see the LICENSE file in the project repository.

Be the First to Experience MCP Now