Neo4j Data Modeling MCP Server: Design Graph Models in Claude, Cursor and Other Apps

• Checks one node definition

  Checks that a single node definition has the right shape and raises an error that says what is wrong. It can return the cleaned node.

  Try: “Check that my Customer node definition is valid.”

  validate_node

• Checks one relationship definition

  Checks that a relationship definition is well formed, including its type and its start and end nodes.

  Try: “Is this PURCHASED relationship between Customer and Product valid?”

  validate_relationship

• Checks a whole data model

  Checks every node and relationship in a model together and reports problems, such as links to nodes that do not exist.

  Try: “Validate the whole data model before we go on.”

  validate_data_model

• Draws the model as a diagram

  Returns Mermaid diagram text for a data model, which Claude Desktop and other apps with Mermaid support can show as a picture.

  Try: “Show my data model as a diagram.”

  get_mermaid_config_str

• Imports a model drawn in Arrows

  Converts JSON exported from the Arrows graph drawing app into a data model.

  Try: “Load this Arrows export as a data model.”

  load_from_arrows_json

• Exports a model for Arrows

  Converts a data model to the JSON format that the Arrows app opens, so you can edit it visually.

  Try: “Export this model so I can open it in Arrows.”

  export_to_arrows_json

• Imports an OWL ontology

  Turns an OWL Turtle ontology into a data model. Classes become nodes, object properties become relationships and data properties become node properties. The conversion loses some detail.

  Try: “Convert this OWL ontology into a Neo4j data model.”

  load_from_owl_turtle

• Exports a model as an OWL ontology

  Writes a data model as OWL Turtle text. Properties on relationships are not kept, because OWL has no place for them.

  Try: “Export the model as an OWL ontology.”

  export_to_owl_turtle

• Exports a model as Python classes

  Returns Python code with Pydantic classes for each node, each relationship and the whole model.

  Try: “Generate Pydantic models for my graph schema.”

  export_to_pydantic_models

• Exports a model for the GraphRAG package

  Converts a data model to the schema dictionary used by the Neo4j GraphRAG Python package.

  Try: “Give me this model as a GraphRAG schema.”

  export_to_neo4j_graphrag_pkg_schema

• Imports a GraphRAG package schema

  Builds a data model from a schema dictionary used by the Neo4j GraphRAG Python package.

  Try: “Load my GraphRAG schema as a data model.”

  load_from_neo4j_graphrag_pkg_schema

• Lists the built-in example models

  Lists seven example models, such as patient journey and supply chain, with a description and the node and relationship counts of each.

  Try: “What example data models do you have?”

  Read · list_example_data_models

• Returns one example model

  Returns a ready-made example model, plus its diagram text, for a domain such as customer 360 or fraud and AML.

  Try: “Show me the supply chain example model.”

  Read · get_example_data_model

• Writes constraint statements

  Generates the Cypher statements that create uniqueness constraints for the key property of each node. It only returns the text and does not run it.

  Try: “Give me the Cypher to create constraints for this model.”

  get_constraints_cypher_queries

• Writes a node load query

  Generates a parameterised Cypher query that loads many records for one node type. You run it elsewhere, for example with the Neo4j Cypher server.

  Try: “Write the Cypher to load Customer records.”

  get_node_cypher_ingest_query

• Writes a relationship load query

  Generates a parameterised Cypher query that loads many records of one relationship type between two node labels. It returns text only.

  Try: “Write the Cypher to load PURCHASED relationships.”

  get_relationship_cypher_ingest_query

Watch out for

• It does not run the Cypher it writes

  The ingest and constraint tools only return query text. Review it before you run it with another server or in the Neo4j browser. github.com

• OWL conversion loses detail

  Converting to and from OWL is lossy, and relationship properties are not kept in OWL. Check the result before you rely on it. github.com

• HTTP mode defaults to localhost

  If you choose http or sse, the server binds 127.0.0.1 unless you set another host. The Docker image sets 0.0.0.0, so publish its port with care. github.com

Security advisories