Neo4j Agent Integrations

Integration guides and working examples for connecting Neo4j to 20+ AI agent platforms and frameworks.

Overview

This repository provides self-contained integration guides showing how to build agents that use Neo4j as their knowledge layer. Each integration demonstrates the same reference implementation adapted to the platform's patterns.

Quick Start

1. Choose a platform from Platform Coverage
2. Navigate to its folder (e.g., langraph/, aws-agentcore/)
3. Follow the README for setup and examples
4. Use the demo database to test

Demo Database

All examples use a read-only Neo4j instance with company and news data:

NEO4J_URI = "neo4j+s://demo.neo4jlabs.com:7687"
NEO4J_USERNAME = "companies"
NEO4J_PASSWORD = "companies"
NEO4J_DATABASE = "companies"

Dataset: 250k entities from Diffbot's knowledge graph

• Organizations, people, locations, industries
• News articles with vector embeddings

Data model:

(Organization)-[:LOCATED_IN]->(Location)
(Organization)-[:IN_INDUSTRY]->(Industry)
(Person)-[:WORKS_FOR]->(Organization)
(Article)-[:MENTIONS]->(Organization)
(Article)-[:HAS_CHUNK]->(Chunk) [vector embeddings]

Reference Implementation

See EXAMPLE_AGENT.md for the canonical "Industry Research Agent" that each platform implements.

The reference agent demonstrates:

• Querying company data from Neo4j
• Vector search over news articles
• Analyzing organizational relationships
• Synthesizing research reports

This provides a consistent pattern for comparing integration approaches across platforms.

Platform Coverage

Cloud Providers

• aws-agentcore - AWS AgentCore Bedrock (native MCP + A2A)
• microsoft-foundry - Azure AI Foundry (native MCP + A2A)
• google-gemini-enterprise - Google Gemini Enterprise (MCP + Vertex AI Extensions)

Vertical Platforms

• databricks-agent-bricks - Databricks (MCP Catalog + Unity Catalog)
• snowflake-cortex - Snowflake Cortex
• servicenow-ai-agents - ServiceNow (MCP + A2A)
• salesforce-agentforce - Salesforce Agentforce

Agent Frameworks

• langgraph - LangGraph (MCP adapters)
• langchain - LangChain (MCP adapters)
• microsoft-agent-framework - Microsoft Agent Framework (AutoGen + Semantic Kernel)
• openai-agents-sdk - OpenAI Agents SDK
• strands-agents - AWS Strands
• google-adk - Google ADK
• crewai - CrewAI
• pydantic-ai - Pydantic AI
• llamaindex - LlamaIndex
• haystack - Haystack
• claude-agent - Claude Agent

Workflow & Orchestration

• ibm-watsonx - IBM watsonx Orchestrate
• n8n - n8n
• langflow - Langflow

Integration Patterns

1. MCP Integration

Use the Neo4j MCP server for standardized tool interface:

from mcp_client import MCPClient

client = MCPClient(url="http://localhost:8000/mcp")
tools = client.list_tools()  # Neo4j query tools available

The MCP server can run:

• Locally via stdio
• As HTTP service (SSE or streamable-http)
• Deployed to cloud (Cloud Run, Lambda, etc.)

2. Direct Integration for Tools

Use Neo4j driver directly for more control:

from neo4j import GraphDatabase

driver = GraphDatabase.driver(
    "neo4j+s://demo.neo4jlabs.com:7687",
    auth=("companies", "companies")
)

def query_company(company_name: str):
    query = """
    MATCH (o:Organization {name: $company})
    RETURN o.name as name,
           [(o)-[:LOCATED_IN]->(loc:Location) | loc.name] as locations,
           [(o)-[:IN_INDUSTRY]->(ind:Industry) | ind.name] as industries
    LIMIT 1
    """
    records, summary, keys = driver.execute_query(
        query,
        company=company_name,
        database_="companies"
    )
    return records[0].data() if records else {}

3. Custom Integrations

Use dedicated extension points of the agent framework to implement a specific integration.

Authentication

Different platforms support different auth mechanisms. See mcp-auth-support.md for detailed comparison.

Common patterns:

• API Keys - Simplest, supported everywhere
• OAuth M2M - Enterprise platforms (AWS, Azure, Databricks, ServiceNow)
• Platform IAM - Cloud-native (AWS IAM, Azure AD, GCP Service Accounts)

Repository Structure

platform-name/
├── README.md              # Integration guide with working examples
├── examples/              # Complete code samples (optional)
└── challenges.md          # Known issues and workarounds (optional)

Each platform README includes:

1. Platform overview and official links
2. Extension points (how to integrate)
3. Authentication setup
4. Complete implementation of the reference agent
5. Challenges and limitations
6. Additional integration opportunities

See INTEGRATION_TEMPLATE.md for the template.

Common Queries

Get company information:

MATCH (o:Organization {name: $company})
RETURN o.name as name,
       [(o)-[:LOCATED_IN]->(loc:Location) | loc.name] as locations,
       [(o)-[:IN_INDUSTRY]->(ind:Industry) | ind.name] as industries,
       [(o)<-[:WORKS_FOR]-(p:Person) | {name: p.name, title: p.title}] as leadership
LIMIT 1

Vector search news articles:

MATCH (o:Organization {name: $company})<-[:MENTIONS]-(a:Article)
MATCH (a)-[:HAS_CHUNK]->(c:Chunk)
CALL db.index.vector.queryNodes('news', 5, $embedding)
YIELD node, score
WHERE node = c
RETURN a.title, a.date, c.text, score
ORDER BY score DESC

Contributing

When adding a new integration:

1. Use INTEGRATION_TEMPLATE.md as starting point
2. Implement the reference agent from EXAMPLE_AGENT.md
3. Include working code examples
4. Document authentication setup clearly
5. Note platform-specific challenges and workarounds
6. Keep examples self-contained (no external dependencies beyond Neo4j MCP server)

Resources

• Neo4j MCP Server: https://github.com/neo4j/mcp
• MCP Specification: https://modelcontextprotocol.io/
• Neo4j Driver Docs: https://neo4j.com/docs/
• Demo Database: neo4j+s://demo.neo4jlabs.com:7687 (companies/companies)