AI Agent Protocols Explained

A key theme coming out of ServiceNow’s Knowledge Conference and Microsoft’s Build Conference has been interoperability of agents. Alongside that, you’ve probably stumbled upon a series of new and confusing acronyms like MCP, A2A, ANP, and ACP.

As always with acronyms, they sound technical, and a bit intimidating but before long they are going to be fully integrated into how we talk about AI Agents. If anything, they’re an indicator of how fast and far platforms are evolving with AI in a very short space of time.

So to help with the learning process, let’s break them down in plain English.

First, let’s go back to our definition of an AI Agent. An AI Agent is able to process a task and use the tools at its disposal to understand the task, collect context from its environment, and act accordingly in pursuit of a specific outcome.

Where the protocols come in is defining how different agents work together so the definition subtly changes to:

An AI Agent is able to process a task and use the tools at its disposal to understand the task, collect context from its environment, collaborate, and act accordingly in pursuit of a specific outcome.

This is very similar to how humans work in the real world. Some workers try to do everything themselves, while others delegate, collaborate, or work with external partners.

Let’s use a really simple example to bring it to life: booking a holiday.

1. MCP: Model Context Protocol

One Agent, All the Tools

This is the simplest approach. Imagine you ask a travel agent to plan your trip. That one person calls the airline, books the hotel, checks the weather, and returns with the full itinerary. No one else is involved.

Pros:

• Simple and centralised

• Easy to manage

• Great for basic tasks

Cons:

• Doesn’t scale well

• Struggles when the task spans multiple systems or domains

This is great for isolated, contained workflows where no deep collaboration across domains or systems is needed.

2. A2A: Agent-to-Agent Protocol

Delegation Within One Organisation

Now imagine your travel agent has a team. One handles flights, another hotels, another activities. Your agent coordinates between them and assembles the full plan.

Each team member (agent) is an expert in their own area, and they talk to each other to get the job done.

Pros:

• Supports specialisation

• Enables collaboration

• Keeps things tidy within one system or department

Cons:

• Works best within one organisation or domain

• Adding cross-company collaboration can get tricky

In cross-departmental automations - e.g. between IT, HR, and Facilities teams - this is likely where early multi-agent use cases will take hold.

3. ANP: Agent Network Protocol

Cross-Company Collaboration, No Middleman

Here, your travel agent doesn’t just work with in-house colleagues - they directly message hotel agents, airline agents, and weather services across different organisations. Acting as the central coordinator between them.

It’s more powerful and flexible, but harder to manage.

Pros:

• Designed for cross-domain automation

• Supports decentralised systems

• Agents can explore, fetch data, and act autonomously

Cons:

• Complex to secure

• Tougher to handle errors and failures

As more AI starts to span multiple external systems (Microsoft Copilot, ServiceNow, Workday, other clouds), this kind of protocol will help to deliver seamless automation across platforms.

4. ACP: Agent Communication Protocol

Structured Messages, Enterprise-Ready

Think of this as the most formal and scalable version. Agents don’t just ‘talk’ - they follow strict rules for messaging like “Request”, “Inform”, “Collaborate”. This creates a stable unified language for agents from different teams, or even different companies, to work together efficiently.

Pros:

• Highly modular and scalable

• Ideal for large, enterprise-wide systems

• Enables safe, structured collaboration with external partners

Cons:

• Needs careful design upfront

• Every message must follow strict rules

This is the ultimate end goal: where AI agents from different vendors and domains can communicate in a standardised, trustworthy way.

Why This Matters

These protocols represent more than just a shift in technology, they signal a change in how enterprise software will function in the future. Instead of workflows that move through static, linear paths, we’ll have swarms of AI agents that negotiate, reason, and collaborate in real time.

And platforms like ServiceNow, Salesforce, and Microsoft aren’t just watching they’re building for it.

So What Should You Do?

• Start tracking agent use cases regardless of the platform they’re built in

• Familiarise yourself with the terminology, when each protocol is appropriate, when it isn’t

• Keep an eye out for new protocols, or evolutions to those that already exist

• Start having a think about where interoperability can start playing a role in your business