The Rise of A2A: Completing the Industrial AI Protocol Stack with OPC UA and MCP

It feels like I was just discussing the rise of the Model Context Protocol (MCP) and how it complements established standards like OPC UA in the Industrial AI landscape. But the truth is, the speed at which AI is advancing today is nothing short of staggering.

Blink and you might miss the next big thing. (Stay tuned—more on that soon. And no, the answer isn’t to sit back and wait for things to mature—that’s not happening anytime soon!)

What’s becoming more and more evident is the immense potential in getting multiple AI agents to work together—coordinating complex tasks without being hindered by application or data silos. This isn’t just a bonus anymore; it’s quickly turning into a must-have for scaling AI effectively. The good news? This realization is fueling a promising shift: open-source, multi-vendor collaboration aimed at building the standards we need.

In my previous post, we looked at how OPC UA offers a secure, essential foundation for accessing and interpreting data from the operational technology (OT) realm. We also explored how MCP serves as a much-needed “USB-C port” for AI—standardizing how individual agents connect to tools and data sources, including that all-important contextualized OPC UA data. Together, they form a powerful duo, bridging the OT/IT gap to make AI integration more seamless.

But as much as MCP empowers individual agents, a gap still lingered. MCP was originally designed for agent-to-tool or agent-to-data interactions—not for the kind of rich, peer-to-peer conversations required when multiple agents need to collaborate dynamically. And in distributed industrial settings, that kind of collaboration is often essential. Forcing agents to interact with each other as if they were just "tools" through MCP felt limiting—it just didn’t capture the depth needed for true multi-agent coordination.

Enter A2A: The Missing Link for Agent Collaboration

Which brings us to the latest piece of the puzzle: the Agent2Agent (A2A) protocol. Introduced in April 2025 by Google—importantly, in collaboration with a powerhouse consortium of over 50 industry partners. This includes enterprise heavyweights like Salesforce, SAP, and ServiceNow, as well as top system integrators such as Accenture, Deloitte, and KPMG.

A2A is purpose-built as an open standard to complement MCP. If MCP is the wrench that lets an agent operate tools—like accessing OPC UA data—then A2A is the common language that lets multiple mechanics (agents) coordinate on a complex job. It zeroes in on enabling secure communication, capability discovery, and task coordination between AI agents, no matter who built them or which framework they run on.

The Industrial AI Protocol Stack Takes Shape

We’re now seeing a potential three-layered stack take shape for robust Industrial AI:

1. OPC UA (Foundation): Secure, semantic data access from OT assets.

2. MCP (Agent Enablement): Standardized interface for agents to access tools and contextual data (including OPC UA data, potentially via data fabrics/UNS).

3. A2A (Agent Coordination): Standardized protocol for these MCP-enabled agents to collaborate and coordinate complex tasks.

A2A builds on well-established web standards like HTTP and JSON-RPC, making it easier to integrate into enterprise IT environments. It introduces key concepts such as Agent Cards for discovering the capabilities of other agents, and Tasks to manage workflows between them—ideal for the long-running, sometimes human-in-the-loop processes typical in industrial settings.

This stack is purpose-built for the complexity of real-world Industrial AI use cases: coordinating robot fleets, managing distributed process control, optimizing multi-actor supply chains, orchestrating predictive maintenance, and enabling dynamic digital twins. For instance, an agent monitoring machine health (using MCP to access OPC UA sensor data) can now coordinate seamlessly via A2A with a maintenance scheduling agent and a parts inventory agent to plan and carry out a repair.

Momentum Toward Interoperability

The sheer number and caliber of companies backing A2A from the outset speaks volumes. It reflects a strong industry consensus that agent interoperability isn’t just a nice-to-have—it’s essential.

This collaborative, open-standard approach—building progressively from OPC UA to MCP to A2A—feels like a step in the right direction. It helps avoid vendor lock-in while smartly building on existing, proven standards.

That said, challenges lie ahead. A2A is still in its early days, the spec continues to evolve, and ensuring robust agent-to-agent security and authorization will require careful design and validation. But the momentum is real.

For those of us navigating the Industrial AI landscape, this layered protocol stack represents the most promising architecture we’ve seen so far for enabling intelligent, collaborative, and scalable automation. It’s one we’ll be watching closely here at ARC Advisory Group.

Links to Previous Blogs

• Bridging Context and Connectivity: How MCP Complements OPC UA in Industrial AI

• Beyond the Hype: Practical Realities of MCP in Industrial AI Deployment

• The Power Trio: MCP, UNS, and CESMII–Orchestrating the Future of Industrial AI

• Contextual Harmony: Industrial AI, Unified Namespaces and the Promise of MCP

• Context Is the Missing Link: The Emergence of the Model Context Protocol in Industrial AI

Engage with ARC Advisory Group

For ARC Advisory Group recommendations for Navigating the AI Wars, Closing the Digital Divide by Embracing Industrial AI, assembling your Industrial-Grade Data Fabric, and governing and guiding major decisions about enterprise, cloud, industrial edge, and AI software, please contact Colin Masson at [email protected] or set up a meeting with me, or my fellow Analysts at ARC Advisory Group.