Neo4j Model Context Protocol (MCP) Servers

The mcp-neo4j-memory server is a Model Context Protocol (MCP) implementation that provides persistent memory capabilities using a Neo4j graph database. It allows LLMs to store and retrieve complex relationships between entities as a long-term knowledge base.

Graph Schema

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

The mcp-neo4j-data-modeling server (mcp-name: io.github.neo4j-contrib/mcp-neo4j-data-modeling) is a Model Context Protocol (MCP) implementation designed for creating, visualizing, and managing Neo4j graph data models.

It allows developers and AI agents to:

• Define nodes, relationships, and properties.
• Design graph database schemas.
• Visualize models interactively using Mermaid.
• Validate data model structures.
• Export/Import models between formats like Arrows JSON, OWL Turtle, and Pydantic models.
• Generate Cypher queries for data ingestion.

Neo4j Labs provides a suite of Model Context Protocol (MCP) servers that allow LLM clients (like Claude Desktop, VS Code, Cursor, Windsurf, or Gemini CLI) to interact with Neo4j databases and services using natural language.

These servers enable tasks such as:

• Querying graph data using natural language to Cypher.
• Managing Neo4j Aura cloud instances.
• Storing and retrieving personal knowledge graph memory.
• Modeling and visualizing graph data structures.

All Neo4j MCP servers support three transport modes:

1. STDIO (Default): Used for local tools and integration with Claude Desktop.
2. SSE: Server-Sent Events, intended for web-based deployments.
3. HTTP: Streamable HTTP, designed for modern web deployments, microservices, and scalable cloud environments.

To run multiple independent instances of the server within the same environment, use the --namespace flag. This prefixes all tools with the provided namespace (e.g., app1-validate_node), allowing multiple configurations to coexist without tool name collisions.

"mcpServers": {
  "neo4j-data-modeling-app1": {
    "command": "uvx",
    "args": [ "mcp-neo4j-data-modeling@0.8.2", "--transport", "stdio", "--namespace", "app1" ]
  },
  "neo4j-data-modeling-app2": {
    "command": "uvx", 
    "args": [ "mcp-neo4j-data-modeling@0.8.2", "--transport", "stdio", "--namespace", "app2" ]
  }
}

The server supports three transport protocols depending on your deployment environment:

• STDIO: Best for local development (e.g., using with Claude Desktop).
• HTTP: Best for remote deployments and microservices.
• SSE (Server-Sent Events): Best for legacy web-based deployments.

To prevent tool name collisions in multi-tenant environments, you can prefix all tool names using the --namespace flag or the NEO4J_NAMESPACE environment variable. For example, if the namespace is myapp, the tool read_graph becomes myapp-read_graph.

"args": [ "mcp-neo4j-memory@0.4.5", "--namespace", "myapp", "--db-url", "..." ]

The server supports namespacing to prefix tool names. This is useful when running multiple instances of the server for different tenants. When a namespace (e.g., myapp) is provided, tools are prefixed accordingly (e.g., myapp-list_instances).

CLI Usage:

mcp-neo4j-aura-manager --client-id <id> --client-secret <secret> --namespace myapp

Environment Variables: Set NEO4J_NAMESPACE to apply the prefix.

export NEO4J_NAMESPACE=myapp
mcp-neo4j-aura-manager

"mcpServers": {
  "neo4j-aura-app1": {
    "command": "uvx",
    "args": [
      "mcp-neo4j-aura-manager@0.4.8",
      "--client-id", "<your-client-id>",
      "--client-secret", "<your-client-secret>",
      "--namespace", "app1"
    ]
  }
}

When using HTTP transport, the server implements DNS Rebinding and CORS protection by default.

Security Best Practices:

• allow_origins: Be specific (e.g., ["https://example.com"]). Never use "*" in production.
• allowed_hosts: Include your actual domain (e.g., ["example.com"]). Include localhost only for development. Never use "*" unless you understand the risks.

# Production Setup
mcp-neo4j-aura-manager --transport http \
  --allowed-hosts "example.com,www.example.com" \
  --allow-origins "https://example.com,https://example.com"

The server supports three distinct transport modes:

• STDIO (default): Used for local tools and Claude Desktop.
• SSE: Server-Sent Events for legacy web-based deployments.
• HTTP: Streamable HTTP for modern web deployments and microservices.

The repository contains several specialized MCP servers:

• mcp-neo4j-cypher: Converts natural language to Cypher queries. It can retrieve database schemas and execute both read and write queries. Requirement: The APOC plugin must be installed and enabled on the Neo4j instance for schema inspection.
• mcp-neo4j-memory: Provides a knowledge graph memory system. It allows storing and retrieving entities and relationships in a local or remote Neo4j instance across different sessions and clients.
• mcp-neo4j-cloud-aura-api: Manages Neo4j Aura cloud services. Use it to create, destroy, find, or scale Aura instances and enable specific features via chat.
• mcp-neo4j-data-modeling: Facilitates interactive graph data modeling and visualization. It supports importing and exporting models from Arrows.app.