One Registry to Rule them All - Sonny Merla, Mauro Luchetti, & Mattia Redaelli, Quantyca

Key Concepts

• Amplify Program: A global, cross-functional initiative at Amplifon to standardize AI adoption, governance, and infrastructure.
• MCP (Model Context Protocol): A standard for connecting AI assistants to systems and data; Amplifon uses a private registry to manage these.
• A2A (Agent-to-Agent): A framework for inter-agent communication using the "Agent Card" standard.
• AI Gateway: A centralized entry point for LLM access, providing unified authentication, budgeting, and auditing.
• Agent Card: A standardized metadata format describing an agent’s identity, capabilities, and endpoints.
• Lineage: The ability to track the relationship between use cases, agents, tools, and models for impact analysis and compliance.

---

1. The Amplify Program: Governance and Structure

Amplifon, a global leader in hearing care, faced "chaos" due to decentralized AI development across 26 countries. To solve this, they launched the Amplify Program in January 2025, structured around:

• Control Tower: A leadership group defining high-level strategy, security, and legal guidelines.
• Committee: A group responsible for executing the strategy and prioritizing use cases at the local and corporate levels.
• Three Pillars:
  • Governance: Ensuring regulatory alignment and organizational awareness.
  • Platform: Certifying infrastructure and standardizing development workflows.
  • Factory: Focusing on the practical rollout of scalable, reusable AI solutions.

2. Technical Architecture: The AI Gateway

To address maintenance, security, and cost, the team implemented an AI Gateway that serves as the central hub for all AI interactions:

• Unified Access: Developers point to a single endpoint for all approved LLMs.
• Security: Integrated with Entra ID for centralized authentication.
• Budgeting: Allows for granular cost tracking (monthly/weekly) per use case, providing developers with real-time visibility into their remaining budget.
• Auditing: Centralized monitoring and analytics for all LLM requests and responses.

3. The Registry System

The platform utilizes three interconnected registries to ensure governance and discoverability:

• MCP Registry: A private catalog of internal and certified public MCP servers. It enriches servers with enterprise metadata: ownership, environment (dev/test/prod), authentication models, cost attribution, and use case linkage.
• A2A Registry: A catalog of available agents based on the Agent Card standard. It enables self-documenting agents that can be discovered and invoked by other agents.
• Use Case Registry: The "connective tissue" that maps agents and tools to specific business use cases, enabling full lineage tracking.

4. Development Framework: Blueprints and CI/CD

To prevent "reinventing the wheel," the team created Template Repositories (Blueprints) for both MCP and A2A development:

• Standardization: Blueprints include boilerplate code, Docker files, and pre-configured authentication/cost-tracking logic.
• Framework Agnostic: The A2A blueprint uses interfaces and ports, allowing teams to use their preferred frameworks (e.g., LangChain, Agno) while maintaining a consistent interface.
• Automated Metadata: Upon deployment, CI/CD pipelines (GitHub Actions) automatically publish the Docker image and update the registry metadata (Agent Card or server.json), ensuring the catalog is always in sync with production.

5. Impact Analysis and Lineage

A critical feature of the platform is the Object Lineage view. By mapping the dependencies between a use case, its agents, the MCP tools they use, and the underlying LLMs, the organization can:

• Perform impact analysis during outages (e.g., identifying which use cases are affected if a specific MCP server fails).
• Maintain a clear audit trail of what AI assets are being used and by whom.

6. Notable Quotes

• "What happens when you have dozens of teams across three continents all building AI agents... each one wiring up their own connections? You get chaos." — Sonny Merla
• "We want to make the life of developers easy to focus on the business logic... avoiding to reinvent the wheel every time." — Sonny Merla

Synthesis and Conclusion

The Amplify platform successfully transitions AI development from a fragmented, high-risk environment to a governed, scalable enterprise model. By combining a centralized AI Gateway for operations with a metadata-driven registry system for governance, Amplifon has created a self-documenting ecosystem. The use of standardized blueprints and automated CI/CD pipelines ensures that developers can focus on business value while the platform handles the complexities of security, compliance, and infrastructure. The result is a transparent, traceable, and efficient AI factory capable of supporting global operations.