Neo4j Query Plugin for Dify

Version: 0.2.0

Type: Tool

Overview

The Neo4j Query Plugin is a secure and efficient integration tool that connects Dify.ai with Neo4j graph databases. It enables powerful graph-based data exploration, querying, and manipulation within Dify workflows and AI agents. The plugin supports both Neo4j Aura cloud instances and locally hosted Neo4j deployments.

Key Features

🔒 Security

• Parameterized queries - Prevents Cypher injection attacks
• Preflight validation - EXPLAIN-based query validation before execution
• Write query protection - Optional write operations with explicit user consent
• Query length limits - Maximum 2000 characters per query
• Resource limits - Configurable maximum record limits (default: 1000)

⚡ Performance

• Singleton driver pattern - Efficient connection reuse
• Streaming results - Memory-efficient record fetching with configurable fetch_size
• Connection pooling - Built-in Neo4j driver connection management
• User agent tracking - Identifies plugin traffic in Neo4j logs

🎯 Flexibility

• Multi-database support - Query different databases within the same instance
• Read/Write control - Granular control over query capabilities
• Both cloud and local - Works with Neo4j Aura and self-hosted instances
• Parameterized queries - Safe value substitution in Cypher queries

Latest Updates (v0.2.0)

• ✨ Write query protection with explicit checkbox and warnings
• ✨ Preflight EXPLAIN checks for syntax validation and query type detection
• ✨ Database parameter for multi-database support
• ✨ Configurable record limits with efficient streaming
• ✨ User agent identification for tracking plugin usage
• ✨ Improved error messages with detailed validation feedback
• ✨ Thread-safe driver management with proper cleanup
• ✨ Better credential validation with specific error messages
• 🔧 Renamed credentials from "Aura" to "Neo4j" for clarity
• 🔧 Optimized memory usage with itertools.islice
• 🔧 Enhanced parameter documentation with examples

Installation

Prerequisites

• Dify instance (self-hosted or cloud)
• Neo4j database (Aura or self-hosted)
• Dify Plugin Development Scaffold (CLI)

Steps

1. Install Dify Plugin CLI

   # Follow instructions from Dify documentation
   # https://docs.dify.ai/plugin-dev-en/0211-getting-started-dify-tool

2. Package the Plugin

   # Navigate to parent folder of dify_neo4j
   ./dify plugin package dify_neo4j

3. Upload to Dify

   • Go to your Dify instance
   • Navigate to Plugins tab
   • Upload the packaged plugin

4. Configure Credentials

   • Neo4j URL: Your connection URL (e.g., neo4j+s://xxxxx.databases.neo4j.io or bolt://localhost:7687)
   • Neo4j Username: Your database username
   • Neo4j Password: Your database password

5. Use in Workflows

   • Add as a tool node in workflows
   • Configure as a tool for AI agents

Usage

Basic Query

{
  "query": "MATCH (p:Person) RETURN p.name, p.age LIMIT 10",
  "parameters": {}
}

Parameterized Query

{
  "query": "MATCH (p:Person {name: $name}) RETURN p",
  "parameters": {"name": "Alice"}
}

Multi-Database Query

{
  "query": "MATCH (n:Node) RETURN n LIMIT 5",
  "database": "movies"
}

Write Query (with permission)

{
  "query": "CREATE (p:Person {name: $name, age: $age}) RETURN p",
  "parameters": {"name": "Bob", "age": 30},
  "allow_write_queries": true
}

Configuration Options

Parameter	Type	Required	Default	Description
query	string	Yes	-	Cypher query to execute
parameters	object	No	{}	Query parameters (JSON object)
database	string	No	null	Target database name
max_records	number	No	1000	Maximum records to return
allow_write_queries	boolean	No	false	Enable write operations

Best Practices

1. Schema Context for LLMs

When using the plugin with AI agents, provide the graph schema as context:

Provide your LLM with:
- Node labels and their properties
- Relationship types
- Constraints and indexes

2. Two-Step Query Approach

For complex queries on large graphs:

1. First query: Extract relevant subgraph
2. Second query: Execute detailed analysis on the subgraph

3. Always Use Parameters

# ✅ GOOD - Parameterized
MATCH (p:Person {name: $name}) RETURN p

# ❌ BAD - String concatenation
MATCH (p:Person {name: 'Alice'}) RETURN p

4. Write Protection

• Keep allow_write_queries disabled by default
• Only enable when necessary
• Review queries before execution
• Test on non-production databases first

Security

See SECURITY.md for comprehensive security documentation including:

• Query validation mechanisms
• Write protection details
• Preflight checks
• Parameter handling
• Best practices

Privacy

This plugin:

• ✅ Sends queries only to your specified Neo4j instance
• ✅ Does not store or transmit data to third parties
• ✅ Credentials are securely stored in Dify
• ✅ All data remains within your Neo4j infrastructure

Contributors

• Nikola Milosevic