Neo4j maintains a family of MCP servers rather than one. The main one turns natural language into Cypher and runs it against your database; the others give an agent graph-backed memory, interactive data modelling, and control of Aura cloud instances.
Bring your own
Claude Sonnet 4.5
Chat with 60+ AI models on the same workflow — switch to a different model mid-conversation and re-run the same prompt, or use Compare mode to put several side-by-side and balance quality vs. cost.
A Neo4j connection URI with username and password. Create a read-only user for query work rather than using neo4j.
How models use it and what it is built for.
mcp-neo4j-cypher is the one most people mean. It reads your database schema — node labels, relationship types, properties — and uses it to generate Cypher for a question asked in plain language, then executes it and returns the result. Schema inspection depends on the APOC plugin being installed, which is worth checking before you assume the server is broken. The reason graph queries suit this pattern well is that the hard part of Cypher is usually the traversal, not the syntax: expressing "customers who bought something also bought by people who churned" is tedious by hand and natural to ask. Three sibling servers round out the set. mcp-neo4j-memory persists entities and relationships as a knowledge graph in Neo4j, which is the same idea as the reference memory server but backed by a real database and queryable with Cypher. mcp-neo4j-data-modeling builds and validates graph data models with Arrows.app import and export. mcp-neo4j-cloud-aura-api manages Aura instances — create, scale, pause, destroy. All of them run over stdio by default and can be switched to SSE or HTTP.
Typical tools an AI model can call. Exact names vary by version.
Taken from the official Neo4j documentation — see neo4j-contrib/mcp-neo4j — official repository for the full reference.
URL format
neo4j://host:7687 · neo4j+s://host:7687 for an encrypted Aura connectionExamples
Local instance
bolt://localhost:7687The default Bolt port on a local install.
Neo4j Aura
neo4j+s://xxxxxxxx.databases.neo4j.ioThe +s scheme is encrypted with certificate verification. Use it for anything managed or remote.
Environment variables
NEO4J_URIrequiredConnection URI for the database.
NEO4J_USERNAMErequiredUsername. Prefer a dedicated read-only user over neo4j.
NEO4J_PASSWORDrequiredPassword for that user.
NEO4J_DATABASEDatabase to query. Defaults to neo4j.
NEO4J_TRANSPORTstdio, sse or http. Defaults to stdio.
Client configuration
uvx — the Cypher server
Schema inspection needs the APOC plugin installed on the database.
{
"mcpServers": {
"neo4j": {
"command": "uvx",
"args": ["mcp-neo4j-cypher@latest"],
"env": {
"NEO4J_URI": "neo4j+s://xxxxxxxx.databases.neo4j.io",
"NEO4J_USERNAME": "readonly_agent",
"NEO4J_PASSWORD": "YOUR_PASSWORD",
"NEO4J_DATABASE": "neo4j"
}
}
}
}Copy any of these into MCP Agent Studio after connecting.
Which customers are connected to more than three flagged accounts within two hops?
Show me the shortest path between these two people and what connects them.
What does the schema look like, and which relationship types are most common?
Find clusters of products frequently bought together and describe each one.
This is not a single-model product: you get the same MCP connection with 60+ models (Claude, GPT, Gemini, DeepSeek, open-weight, and more), you can switch mid-conversation, and you can open Compare mode to run the same prompt against multiple models at once. The card above is a suggested starting point for this server — not the only choice.
Default pick for Neo4j
Claude Sonnet 4.5
Cypher traversals are where weaker models produce syntactically valid queries with the wrong direction or depth. Sonnet 4.5 reads the schema first and gets the pattern right.
Listing tools proves the server is reachable, not that a model can work with it. Evals go further: they read every tool on the server, write a test suite from its real schemas, and run it — code decides pass/fail on the responses (schema conformance, error codes, pagination, result caps) while a scoring model grades plain-English tasks driven through the tools.
Get a pass/fail report per tool with the evidence behind each verdict — and replay the same suite after every schema change. Destructive tools are excluded from the run.
Run evalsOpen MCP Agent Studio with the connection pre-filled. Add your token, pick any of 60+ models, and start chatting — no install required.
Open Agent StudioCommon questions about connecting, scoping and using it safely.
It is a family of MCP servers maintained under neo4j-contrib. mcp-neo4j-cypher generates and runs Cypher from natural language; mcp-neo4j-memory stores a knowledge graph in Neo4j; mcp-neo4j-data-modeling builds graph models; mcp-neo4j-cloud-aura-api manages Aura instances.
For schema inspection, yes. get_neo4j_schema relies on APOC, and without it the agent has to guess at labels and relationship types — which produces exactly the confident, wrong Cypher you installed the server to avoid. Aura includes APOC by default.
write_neo4j_cypher accepts any write query, so yes, unless you stop it. Connect with a user granted only read privileges for query work, and keep the write path on a separate connection you enable deliberately. Neo4j role-based access control is the boundary that actually holds.
The reference memory server keeps a knowledge graph in a local JSON file — simple, portable, no infrastructure. mcp-neo4j-memory keeps the same shape in a real graph database, so the memory is queryable with Cypher, shareable across agents and able to grow past what fits in a file.
Because relationship questions are where SQL gets painful and Cypher does not. Multi-hop traversals, shortest paths and cluster detection are one-line patterns in Cypher, which makes them a good fit for exactly the open-ended questions people ask an agent.