Neo4j MCP Server

Interact with Neo4j graph databases through MCP: run Cypher queries, inspect schema, manage knowledge graph memory, and model graph data.

Database by Neo4j Labs (neo4j-contrib) Basic Auth active

GitHub Docs

Overview

The Neo4j Labs MCP servers are a collection of Model Context Protocol servers maintained by the Neo4j Field GenAI team that let MCP-compatible clients (Claude Desktop, Cursor, VS Code) interact with Neo4j graph databases and related tooling. The repository at neo4j-contrib/mcp-neo4j bundles four servers covering different graph workflows.

The four servers are: mcp-neo4j-cypher for executing read/write Cypher queries and inspecting schema (requires APOC), mcp-neo4j-memory for persistent knowledge graph storage of entities, relations, and observations, mcp-neo4j-cloud-aura-api for managing Neo4j Aura cloud instances, and mcp-neo4j-data-modeling for designing, validating, and exporting graph data models compatible with Arrows.app and Neo4j GraphRAG.

All servers are Python-based, distributable via uvx or pip, and support STDIO (default), SSE, and HTTP transports. The Labs collection is actively maintained but is marked as experimental, with no SLA or backwards-compatibility guarantees. Neo4j also publishes a separate official product server at github.com/neo4j/mcp focused specifically on Cypher querying and schema discovery.

Tools

Tool	Description
`read_neo4j_cypher`	Execute Cypher read queries against the database.
`write_neo4j_cypher`	Execute Cypher write/update queries; can be disabled with --read-only flag.
`get_neo4j_schema`	Retrieve node labels, properties (name/type), and relationships. Requires the APOC plugin.
`read_graph`	Read the entire knowledge graph from the memory server.
`search_nodes`	Search the knowledge graph for nodes matching a query string.
`find_nodes`	Find specific nodes by name.
`create_entities`	Create new entities in the knowledge graph with name, type, and observations.
`delete_entities`	Delete entities and their associated relations.
`create_relations`	Create relations between entities (source, target, relation type).
`delete_relations`	Delete relations from the graph.
`add_observations`	Add observations to existing entities.
`delete_observations`	Delete observations from entities.
`validate_node`	Validate a single node structure in a data model.
`validate_relationship`	Validate a single relationship structure.
`validate_data_model`	Validate the full data model structure.
`get_mermaid_config_str`	Generate a Mermaid diagram for data model visualization.
`load_from_arrows_json`	Import a data model from Arrows.app JSON format.
`export_to_arrows_json`	Export the current data model to Arrows.app JSON.
`export_to_pydantic_models`	Export the data model as Python Pydantic classes.
`get_node_cypher_ingest_query`	Generate a Cypher ingestion query for a node.
`get_relationship_cypher_ingest_query`	Generate a Cypher ingestion query for a relationship.
`get_constraints_cypher_queries`	Generate Cypher constraint-creation queries for a data model.

Setup Guide

Prerequisites

A running Neo4j instance (Aura cloud or self-managed)
For get_neo4j_schema: the APOC plugin installed and enabled
uv/uvx installed (recommended) or Python with pip

Install

The servers are published to PyPI and run via uvx:

uvx mcp-neo4j-cypher@0.6.0 --transport stdio

Alternative servers in the repo:

uvx mcp-neo4j-memory
uvx mcp-neo4j-data-modeling
uvx mcp-neo4j-cloud-aura-api

Claude Desktop config

Add to claude_desktop_config.json:

{
  "mcpServers": {
    "neo4j-database": {
      "command": "uvx",
      "args": ["mcp-neo4j-cypher@0.6.0", "--transport", "stdio"],
      "env": {
        "NEO4J_URI": "bolt://localhost:7687",
        "NEO4J_USERNAME": "neo4j",
        "NEO4J_PASSWORD": "<your-password>",
        "NEO4J_DATABASE": "neo4j"
      }
    }
  }
}

Environment variables

NEO4J_URI: Bolt connection URI (e.g. bolt://localhost:7687 or neo4j+s://xxx.databases.neo4j.io)
NEO4J_USERNAME, NEO4J_PASSWORD: database credentials
NEO4J_DATABASE: target database (default neo4j)
NEO4J_READ_ONLY: set to true to disable write tools
NEO4J_TRANSPORT, NEO4J_MCP_SERVER_HOST, NEO4J_MCP_SERVER_PORT, NEO4J_MCP_SERVER_PATH: configure HTTP/SSE transport for remote deployments

Use Cases

Ask natural-language questions about a Neo4j graph and have the agent generate, run, and summarize Cypher queries.
Inspect an unfamiliar Neo4j database's schema (labels, relationship types, property keys) before writing application code.
Give an AI assistant persistent, structured memory across sessions using mcp-neo4j-memory as a knowledge graph.
Design a graph data model interactively, validate it, and export ingestion Cypher or Pydantic classes for code generation.
Provision and scale Neo4j Aura cloud instances from an agent workflow via mcp-neo4j-cloud-aura-api.

Example Prompts

"What node labels and relationship types exist in my Neo4j database?"
"Find all Person nodes connected to Company 'Acme' through WORKS_AT and list their titles."
"Create a node for product SKU-1234 and link it to category 'Electronics'."
"Remember that Alice is the project lead for the Phoenix initiative and works with Bob and Carol."
"Validate this data model and generate the Cypher constraints and Pydantic classes for it."

Pros

Maintained by Neo4j's Field GenAI team, so coverage tracks the product (Aura, GraphRAG, Arrows.app).
Four complementary servers cover querying, memory, modeling, and cloud management in one repo.
Supports STDIO, SSE, and HTTP transports, enabling local and containerized cloud deployments.
Read-only mode via --read-only flag protects production databases from accidental writes.

Limitations

Marked experimental: no SLA or backwards-compatibility guarantees between versions.
Schema discovery in mcp-neo4j-cypher requires the APOC plugin, which is not enabled by default on all instances.
No fine-grained authorization; the server inherits whatever permissions the configured Neo4j user has.

Alternatives

neo4j/mcp: Neo4j's separate official product MCP server, focused on Cypher and schema.
neo4j-field/mcp-neo4j-graphrag: Specialized server adding vector search, fulltext search, and search-augmented Cypher for GraphRAG.
Memgraph MCP server: alternative property-graph database with its own MCP integration.