Enterprise Registry Unifies MCP & A2A Agents at Scale

Build private MCP and A2A registries enriched with enterprise metadata to enable discovery, governance, lineage, and standardized deployment across global teams building AI agents.

Tackle Decentralized AI Chaos with Unified Registries

Global teams across 26 countries, 20,000+ employees, and 10,000 stores at Amplifon faced reinvented security, fragmented infrastructure, and untraceable AI agents. Their Amplify program (launched Jan 2025) counters this via an AI gateway and three interconnected registries: MCP (tools/functions for LLMs), A2A (agent-to-agent), and Use Case. The gateway provides unified model access, Intra ID authentication, per-use-case budgeting (monthly/weekly), and centralized auditing/monitoring. MCP registry extends the open-source spec with internal/custom servers plus certified public ones, adding metadata like ownership (team/project), environments (dev/test/prod), auth models, cost attribution (tied to gateway budgets), and use case linkages. This enables impact analysis: trace server usage, audit trails, and lineage across assets. A2A registry uses agent cards (identity, endpoint, capabilities, modalities, auth) auto-published via CI/CD on deployment, making agents self-discovering and interoperable. Use Case registry maps agents/tools/models to business contexts, tracking status, versions, lifecycle history, serving systems, and inter-use-case impacts for regulatory compliance and maintenance (e.g., LLM disruptions).

Standardize Development with Production-Ready Blueprints

Provide GitHub template repos for MCP and A2A servers using FastAPI for consistent exposure, Dockerfiles, package management, built-in auth, and cost tracking. Integrate Langfuse for observability (tracing, evals, performance). A2A blueprint is framework-agnostic (LangChain, AutoGen, etc.) via ports/interfaces, letting teams focus on business logic while ensuring uniform interfaces. Wizards generate server.json (MCP) or agent cards (A2A) from forms, with JSON previews. Inspectors validate MCP servers or A2A compatibility in new tabs. Link repos to DevOps: tag a branch, GitHub Actions build/push Docker images to artifact repo and auto-publish metadata to registry backend.

Enable Runtime Discovery and Secure Routing

Agents discover MCP/A2A services dynamically: route via AI gateway proxies (MCP/A2A proxies) that query registries for backend URLs, then forward with auth headers. This hides fragmentation—developers point to unified endpoints without knowing internals. Demo dashboard catalogs 6+ entities (use cases, MCPs, A2As), shows lineage graphs (e.g., 'Ticket Optimization' use case links to agents/models), and supports CRUD for use cases/assets.

Deliver Governance, Traceability, and Scalability

Outcome: Full catalog visibility across teams/continents; traceability from use cases to agents/tools/models; standardized blueprints/CI/CD reduce reinvention; prompt maintenance (e.g., model outages trigger targeted fixes via lineage). Production deployment imminent, with ongoing expansions. Contact speakers for blueprints or discussions.

Video description
As internal MCP servers and A2A agents explode in number, discovery and governance become critical challenges for production-grade AI systems. We'll demonstrate how we built an enterprise infrastructure to index MCP servers and A2A agents, and link them to relevant use cases. We'll show how moving from a fragmented environment to a searchable, metadata-rich registry transformed a chaotic development cycle into a standardized, scalable deployment process. In this talk, we'll cover: - How we developed an internal private company MCP registry based on the open source specification - How we defined an A2A registry based on agent cards - How we achieved agent runtime discovery using an MCP server that exposes company A2A agents - How we linked A2A agent and MCP server template repositories to DevOps processes Mauro Luchetti - AI CoE Manager, Quantyca I work as an AI Engineer and CoE Manager at Quantyca, where I focus on artificial intelligence solutions, data engineering, and cloud architectures, drawing on nearly 8 years of professional experience in the field. Over the years I've had the opportunity to work on projects involving generative AI, machine learning, data governance and data management, trying to combine hands-on technical skills with a broader strategic perspective. I enjoy sharing what I've learned with the teams I work with, contributing to collective growth in modern AI engineering practices. Socials: https://www.quantyca.it/ Slides: https://quantyca-my.sharepoint.com/:b:/g/personal/mauro_luchetti_quantyca_it/IQBUCcMBzsAfSZtJXrCdaqV0AaUyDhifxP360fqCUupyaGc?e=S6ytoA

Summarized by x-ai/grok-4.1-fast via openrouter

7146 input / 1411 output tokens in 14889ms

© 2026 Edge