# neo4j-contrib/mcp-neo4j: Four Neo4j MCP Servers for Claude, Cursor and Gemini CLI

> The mcp-neo4j repository ships four Labs-stage MCP servers that let an AI assistant read a graph schema, run Cypher, keep a memory graph and manage Aura instances. They are MIT-licensed, containerized, and explicitly not covered by Neo4j product support.

**neo4j-contrib/mcp-neo4j** — Neo4j Labs Model Context Protocol servers

- Repository: https://github.com/neo4j-contrib/mcp-neo4j
- Website: https://neo4j.com/developer/genai-ecosystem/model-context-protocol-mcp/
- Stars: 985 · Forks: 260
- Language: Python
- License: MIT
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/neo4j-contrib-mcp-neo4j

## What mcp-neo4j actually provides, and who is meant to use it

The repository is not one server. It is four, each published as its own package: mcp-neo4j-cypher for natural language to Cypher, mcp-neo4j-memory for a personal knowledge graph, mcp-neo4j-cloud-aura-api for Aura instance management, and mcp-neo4j-data-modeling for creating and visualizing graph data models with Arrows.app import and export. A fifth artifact, mcp-gemini-ext, appears in the release list, which matches the gemini-extension.json and gemini-extension/ entries at the top level of the repository.

The target user is someone who already has a Neo4j database and an MCP client such as Claude Desktop, VS Code, Cursor, Windsurf or Gemini CLI. The README's own examples are the clearest statement of intent: asking what is in a graph, rendering a chart of top products by frequency, listing instances, creating an Aura Professional instance with 4GB and Graph Data Science enabled, or storing the fact that you worked on the MCP servers with two named colleagues. Three of those four are read-oriented or exploratory. The write paths exist, but the framing is assistant-driven exploration, not application integration.

The important caveat sits in the first paragraph. These servers come from the Neo4j Labs program, are developed by the Neo4j Field GenAI team, and are not supported by the Neo4j product team. The README says they are actively developed and maintained, but also that there are no SLAs or guarantees around backwards compatibility and deprecation. If you need the supported path, the README points to a different repository, neo4j/mcp.

## How the four servers divide the work

mcp-neo4j-cypher is the one with a hard external dependency. It gets the database schema for a configured database and executes generated read and write Cypher against it. Schema inspection requires the APOC plugin to be installed and enabled on the Neo4j instance. That single requirement rules it out for locked-down managed instances where you cannot install plugins, and it is the first thing to check before blaming the server for a failed schema call.

mcp-neo4j-memory stores and retrieves entities and relationships from a personal knowledge graph in a local or remote Neo4j instance. The design point is persistence across sessions, conversations and clients, which is a different problem from query generation: the assistant is not asking your database questions, it is writing its own notes into one.

mcp-neo4j-cloud-aura-api manages Neo4j Aura instances from the assistant chat. The README lists creating and destroying instances, finding instances by name, scaling up and down, and enabling features. Destroy is in that list, so this server carries the sharpest blast radius of the four and deserves the narrowest credentials.

mcp-neo4j-data-modeling handles model creation, validation and visualization, with import and export to Arrows.app. It is the only one of the four that does not obviously depend on a live database for its core job.

## Transport modes: STDIO by default, HTTP for deployment

All four servers support three transport modes. STDIO is the default and is described as the mode for local tools and Claude Desktop integration. SSE is listed for web-based deployments. HTTP is described as streamable HTTP for modern web deployments and microservices. The default matters: if you launch a server without arguments under a client that expects a socket, you get STDIO and the client will fail to connect rather than fall back.

The repository states that all servers are containerized and ready for cloud deployment on AWS ECS Fargate and Azure Container Apps, and that HTTP transport is the mode designed for auto-scaling and load balancing. A separate README-Cloud.md is referenced as the complete cloud deployment guide, covering ECS Fargate with an Application Load Balancer, Azure Container Apps, configuration and monitoring guidance, and client integration examples. The main README does not reproduce those steps, so the cloud path is only as good as that second document.

One thing the README does not document is authentication for the HTTP transport. It shows host, port and path configuration, and nothing about tokens or TLS. Treat an HTTP-mode server as something to place behind your own access control until you have read README-Cloud.md and confirmed what it says about security.

## Installing mcp-neo4j-cypher and running a first query

The README does not give a pip or uvx install line for the individual servers, so the reliable entry point is the transport configuration it does document. The simplest first run is HTTP mode with an explicit host, port and path, which makes it obvious whether the process started and what address it bound to.

```bash
mcp-neo4j-cypher --transport http --host 127.0.0.1 --port 8080 --path /api/mcp/
```

If the process starts, it is listening on 127.0.0.1:8080 at the path /api/mcp/. If it exits immediately, the usual cause is a missing or unreachable Neo4j connection, not the transport flags.

The same configuration is available through environment variables, which is the form you want in a container or a systemd unit rather than a long command line.

```bash
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/
mcp-neo4j-cypher
```

After the server is reachable, register it with your MCP client and ask a schema question such as "What is in this graph?". That question exercises the APOC-dependent schema path, so a failure here tells you the plugin is missing before you spend time on query generation. The README's other worked examples, listing Aura instances or storing a fact in the memory graph, belong to the other three servers and need those servers configured instead.

## Where mcp-neo4j is the wrong tool

The Labs status is not a formality. The README states plainly that there are no guarantees around backwards compatibility and deprecation. A tool name, a parameter or an environment variable can change between releases, and the release list shows versioned packages moving independently: mcp-neo4j-cypher at v0.6.0, mcp-neo4j-memory at v0.4.5, mcp-gemini-ext at v1.0.1. If your integration pins a specific server behaviour, you own that pin.

The Cypher server is also the wrong tool when the assistant should not have write access. The README describes it as executing generated read and write Cypher. No read-only mode is documented, so the safe deployment is a database account whose privileges you have narrowed yourself, not a default admin connection. Similarly, mcp-neo4j-cloud-aura-api can destroy instances. An assistant that misreads a sentence about cleanup can end an instance you wanted.

Finally, if you need the officially supported Neo4j MCP server, this repository is not it. The README directs you to neo4j/mcp for that. Choosing mcp-neo4j means accepting experimental features in exchange for the Field GenAI team's faster iteration.

## The alternative: the official Neo4j MCP server

The README names one alternative directly: the official product Neo4j MCP server at github.com/neo4j/mcp, described as the place to look if you want the official product rather than the Labs servers. The difference is not feature count, it is the support contract. The Labs servers are built by the Neo4j Field GenAI team, welcome community contributions, and are explicitly outside Neo4j product team support. The official server sits inside the product organisation.

That distinction should drive the decision more than any capability list. If you are prototyping an assistant against a development database, the Labs servers give you four distinct capabilities, including Aura management and data modeling, that the README presents as a coherent set. If you are shipping something that a support contract has to cover, the Labs servers are the wrong side of that line and the README says so itself.

Within the repository, the four servers are also alternatives to each other. If all you want is a memory graph that survives across chat sessions, mcp-neo4j-memory is the relevant package and you do not need the Cypher server or its APOC requirement at all.

## Licence, maintenance and upgrade cost

The repository is MIT licensed, with LICENSE.txt at the top level. MIT is permissive: you can use, modify and redistribute the code, including commercially, provided the copyright notice and permission notice are retained. That is a statement about the licence text, not legal advice, and it says nothing about the Neo4j database itself, which has its own licensing and is a separate question from this repository.

The maintenance picture is mixed and worth stating precisely. The last push to the default branch was on 2026-09-09. The most recent release in the list is mcp-neo4j-cypher-v0.6.0 from 2026-04-10, followed by mcp-gemini-ext-v1.0.1 on 2026-03-18 and mcp-neo4j-memory-v0.4.5 on 2026-02-23. So the repository sees commits more recently than its last tagged release, and the three packages are versioned separately rather than released together.

The upgrade cost follows from that versioning. Because each server is its own package with its own version, upgrading the Cypher server does not move the memory server, and a client configuration that references all four has four independent version pins to track. Combined with the stated absence of backward-compatibility guarantees, the practical approach is to pin versions, read the release notes for the specific package you use, and expect that a minor bump may require a configuration change.

## Conclusion

Adopt mcp-neo4j if you already run Neo4j and want an assistant to inspect the schema and execute Cypher, or if you want a persistent memory graph across chat sessions. Do not adopt it if you need a supported product with backward-compatibility guarantees, since the README states there is no SLA and the official Neo4j MCP server lives in a separate repository. Before wiring it into a shared cluster, verify that APOC is installed and enabled, and confirm whether your client speaks STDIO or HTTP.

## FAQ

### What is mcp-neo4j?

It is a Neo4j Labs repository of Model Context Protocol servers that connect MCP clients such as Claude Desktop, VS Code, Cursor, Windsurf and Gemini CLI to Neo4j. It contains four servers: cypher, memory, cloud-aura-api and data-modeling.

### How do I install and run an mcp-neo4j server in Docker or HTTP mode?

The README documents running a server in HTTP mode with the --transport http flag and --host, --port and --path options, or with the NEO4J_TRANSPORT, NEO4J_MCP_SERVER_HOST, NEO4J_MCP_SERVER_PORT and NEO4J_MCP_SERVER_PATH environment variables. It also states that all servers are containerized and ready for cloud deployment, with README-Cloud.md covering AWS ECS Fargate and Azure Container Apps.

### Does mcp-neo4j-cypher need the APOC plugin?

Yes. The README states that schema inspection requires the APOC plugin to be installed and enabled on the Neo4j instance.

### What can the mcp-neo4j memory server do?

It stores and retrieves entities and relationships from a personal knowledge graph in a local or remote Neo4j instance. The README notes that the information can be accessed over different sessions, conversations and clients.

### Is mcp-neo4j the official Neo4j MCP server?

No. The README describes these as Neo4j Labs servers developed by the Neo4j Field GenAI team and not supported by the Neo4j product team, and points to github.com/neo4j/mcp for the official product server.

## Sources

- [License: MIT](https://github.com/neo4j-contrib/mcp-neo4j/blob/main/LICENSE)
- [neo4j-contrib/mcp-neo4j on GitHub](https://github.com/neo4j-contrib/mcp-neo4j)
- [Project website](https://neo4j.com/developer/genai-ecosystem/model-context-protocol-mcp/)
- [README](https://github.com/neo4j-contrib/mcp-neo4j/blob/main/README.md)
- [Releases](https://github.com/neo4j-contrib/mcp-neo4j/releases)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/neo4j-contrib-mcp-neo4j