Skip to main content

How to block unsafe Cypher queries in LangChain Neo4jGraph

To block unsafe Cypher queries in LangChain Neo4jGraph, guard the driver of Neo4jGraph with guard(graph._driver). Blocked queries raise GuardError and are never sent.

Install​

pip install neo4j-guard langchain-neo4j langchain-anthropic

Guard the driver​

Guard the driver after creating Neo4jGraph. Neo4jGraph loads the schema when it is created, using APOC procedures that read mode blocks. Calling graph.refresh_schema() after guard() raises GuardError.

from langchain_neo4j import Neo4jGraph
from neo4j_guard import GuardError, guard

graph = Neo4jGraph(url="neo4j://localhost:7687", username="neo4j", password="password")
guard(graph._driver)

try:
    graph.query("MATCH (p:Person) DETACH DELETE p")
except GuardError as error:
    print(error)  # DELETE is not allowed in read mode

Use the graph in GraphCypherQAChain​

GraphCypherQAChain runs the LLM's queries through graph.query(), so they go through the guarded driver. The chain requires allow_dangerous_requests=True.

from langchain_anthropic import ChatAnthropic
from langchain_neo4j import GraphCypherQAChain, Neo4jGraph
from neo4j_guard import GuardError, guard

graph = Neo4jGraph(url="neo4j://localhost:7687", username="neo4j", password="password")
guard(graph._driver)

chain = GraphCypherQAChain.from_llm(
    ChatAnthropic(model="claude-opus-5-5"),
    graph=graph,
    allow_dangerous_requests=True,
)

try:
    print(chain.invoke({"query": "Delete every person named Ada"})["result"])
except GuardError as error:
    print(error)  # DELETE is not allowed in read mode