Summary
A Cypher query gets written against the graph that is actually there.
What it takes off you is the guess about labels and relationship types: the schema call returns them with property types and which ones are indexed, so the query that follows uses names that exist. Reads and writes being separate calls means a query that was meant to read cannot quietly write.
What it is
Three calls over a Neo4j database: one returns the schema, and reads and writes are split into two Cypher calls rather than one.
What you get
- The schema: nodes, their properties with types and index flags, and the relationships between them
- A read Cypher query with its own parameters
- A write Cypher query with its own parameters
- A sample size on the schema call, for tuning how long the inspection takes on a large graph
Requirements
Uv, to run `uvx mcp-neo4j-cypher`, a Neo4j database to point at, and APOC installed on it — the schema call inspects through APOC.
Setup effort
One command — uvx mcp-neo4j-cypher