On 17–18 September 2026, Amsterdam hosted AGNTCon + MCPCon Europe 2026, the European gathering of the Agentic AI Foundation community and one of the most technical events of the year on AI agents and the Model Context Protocol (MCP). Omnys was there as a Gold Sponsor, with a large team on the ground: some staffing our booth, others spread across the keynotes and breakout sessions to bring home the latest developments and best practices around agentic systems. At our booth, meanwhile, we talked about two products: our agentic enterprise platform, Omnys AI Platform, and our AI voice agent, Omnys AI Connect.
From protocol to infrastructure: MCP grows up
The through-line of the two days was clear: MCP has moved from an interesting idea to shared infrastructure. In the keynote “MCP and the Era of Connectivity,” David Soria Parra, co-creator of MCP at Anthropic, described the protocol as the connective tissue between models, tools and data; and the rest of the agenda showed a protocol maturing fast: stateless MCP transports heading into production, conformance testing against the latest version of the spec, discovery of MCP servers at organizational scale, and the A2A (agent-to-agent) protocol reaching a stable release.
The Internet of Agents needs a control plane
The second big theme was the conversation that follows every promising technology: how do you run it safely in production? After sitting in on several technical sessions on agent identity, delegated authorization, fine-grained permissions and identity brokers, the conclusion was clear: autonomy is only useful when it is governed. And nowhere is this more evident than in agentic payments and agentic commerce. In the talk “Agents Can Pay. Can They Prove It?”, Diego Zuluaga, Head of Open Mobile Hub (Linux Foundation), put the uncomfortable question bluntly: it is easy to demo an agent that pays or verifies a user, but how does an AI agent request a real, government- or bank-grade credential from someone’s device — on any phone, any wallet, Android or iOS — and prove who actually authorized the transaction? The session ran the full chain live, with an MCP server at its centre, showing the EUDI Wallet architecture working inside an agent a full 18 months before Europe’s December 2026 mandate.
The message for anyone building agents that touch money or identity: the payment is the easy part; the credential, the consent and the audit trail are the product. Who is the agent, what is it allowed to do, and proving after the fact that a real person authorized it: that is the gate between a promising pilot and a system that can go into production.
These themes are the very principles we had already designed our products around.
Omnys AI Platform is an agentic enterprise platform that connects to the systems already present in the organization, such as SQL databases, REST APIs, ERPs, MES, SaaS, and reasons over your proprietary data through MCP connectors, executing multi-step workflows with full auditability. It runs natively on AWS, inside your own account, so your data never leaves your perimeter. The “control plane” conversation we heard in Amsterdam is, in effect, a description of how we believe enterprise AI should be built: governed, traceable and answerable.
Omnys AI Connect, our AI voice agent, applies the same philosophy to customer conversations. It answers calls in natural language, checks its answers against your own documents (RAG) and, this is where MCP comes in, queries your live systems as tools during the call to check a real order, invoice or appointment, then hands the
call to a person with full context only when needed. It is a full agent, not a chatbot with a voice, and like the platform it runs in the company’s own AWS account, with every conversation recorded and reviewable.
Open standards, not new lock-in
One last signal struck us: the ecosystem is moving toward open standards, not toward yet another walled garden. MCP to connect tools and data, A2A to let agents talk to one another, testable and conformant specs: the people building today want to be able to change model, tool or vendor without rewriting everything. That is the same freedom we give our customers: Omnys products run in the customers’ own AWS accounts, integrate with the systems they already have, and never hide what they do behind a black box. In a market wary of lock-in, all of this is a guarantee of “open and verifiable.”
What we’re taking back
Three lessons will shape our roadmap. First, MCP is becoming the default way agents reach tools and data, and we were right to build on it. Second, governance, identity and auditability are the real key to enterprise adoption. Third, evaluation and observability deserve a front-row role; “confidently wrong” agents are the enemy of trust. None of this surprises us, but hearing it confirmed by the people building this sector’s infrastructure is the best kind of validation.
Thank you to the Agentic AI Foundation, the organizers, the speakers and everyone who stopped by our stand. If you didn’t catch us in Amsterdam and want to see how Omnys AI Platform or Omnys AI Connect could work on your own data and systems, let’s talk: write to us at info@omnys.com or get in touch here.