Google Gemini Enterprise + Neo4j Integration
This repository contains a suite of reference implementations for connecting Neo4j Graph Databases to Google Gemini Enterprise.
Using the Agent-to-Agent (A2A) protocol and the Google Agent Development Kit (ADK), these patterns allow organizations to move beyond simple RAG and into high-reasoning GraphRAG workflows.
📂 Repository Structure
Choose the architecture that best fits your deployment needs:
1. a2a-mcp-wrapper
Focus: Modularity and Official Tooling. How it works: This version wraps the Official Neo4j Model Context Protocol (MCP) server. It uses a decoupled architecture where the reasoning agent and the database tool service scale independently on Cloud Run. Best for: Developers who want to leverage the standard MCP ecosystem and prefer a clean separation between the LLM planner and the database driver.
2. a2a-direct-service
Focus: Internal Multi-Tenancy and Simplicity.
How it works: A streamlined version of our agent that uses Native Google Workspace Authentication. It features a /setup portal where users can map their own Neo4j credentials to their Google email address.
Best for: Internal enterprise tools where different teams need to access their specific graphs using their existing corporate Google identities.
3. a2a-ge-marketplace
Focus: Agent-as-a-Service (AaaS) & Commercialization. How it works: The full-featured Google Cloud Marketplace implementation. Includes automated onboarding via Pub/Sub, Dynamic Client Registration (DCR), custom OAuth 2.0 flows, and automated token-based billing. Best for: Partners and SaaS providers looking to publish a "Paid" or "Freemium" agent directly on the Google Cloud Marketplace.
⚖️ Comparison at a Glance
| Feature | MCP Wrapper | Direct Service | Marketplace (AaaS) |
|---|---|---|---|
Multi-Tenancy |
Optional |
Email-Based |
Entitlement-Based |
Authentication |
Google OAuth |
Google OAuth |
Custom OAuth + DCR |
Provisioning |
Manual |
Self-Service (/setup) |
Automated (Pub/Sub) |
Billing |
N/A |
Usage Tracking |
GCP Marketplace Lifecycle |
Architecture |
Decoupled (Agent + MCP) |
Monolithic (All-in-one) |
Monolithic (All-in-one) |
🛠️ Common Prerequisites
Regardless of the version you choose, you will generally need:
Google Cloud Project with Vertex AI and Cloud Run APIs enabled. Secret Manager to handle your Neo4j credentials and internal keys. A Neo4j Instance (AuraDB, Self-Hosted, or Docker). OAuth 2.0 Client IDs configured in the Google Cloud Console.