MCP, ACP, and Agent2Agent set standards for scalable AI results

Open protocols aimed at standardizing how AI systems connect, communicate, and absorb context are providing much needed maturity to an AI market that sees IT leaders anxious to pivot from experimentation to practical solutions.

Three protocols in particular — Model Context Protocol (MCP), Agent Communication Protocol (ACP), and Agent2Agent — show promise for helping IT leaders put two-plus years of failed proof-of-concept projects behind them, opening a new era of measurable AI progress, experts contend.

“In an era where AI is becoming a cornerstone of enterprise strategy, standardization efforts are not merely technical footnotes — they represent the infrastructure of our AI-powered future,” says Zach Evans, CTO at healthcare AI firm Xsolis. “These protocols enable systems to communicate seamlessly across organizational boundaries.”

While other open protocols may still emerge, standardization efforts will likely increase AI adoption rates, Evans says, especially when they facilitate connection to models and agents from multiple vendors — a key issue for the AI agent management challenge IT leaders will soon face.

“When different AI systems are able to more readily interact with one another, history just shows us that that drives adoption,” Evans says. “It’s not just one vendor sitting out there operating in a vacuum. You have the ability for those vendors and those solutions to be able to interact with one another and create smart handoffs.”

The new protocols will enable IT teams to seamlessly connect diverse AI agents and to reduce the cost and complexity of AI integrations, adds Gary Lerhaupt, vice president of product architecture at Salesforce.

“Without standardized protocols, companies will not be able to reap the maximum value from digital labor, or will be forced to build interoperability capabilities themselves, increasing technical debt,” he says.

Protocols are also essential for AI security and scalability, because they will enable AI agents to validate each other, exchange data, and coordinate complex workflows, Lerhaupt adds.

“The industry can build more robust and trustworthy multi-agent systems that integrate with existing infrastructure, encouraging innovation and collaboration instead of isolated, fragmented point solutions,” he says. “For CIOs and CAIOs, this translates to greater flexibility, improved security, and the ability to drive more strategic and efficient AI initiatives across their technology landscape.”

What is MCP?

Model Context Protocol, released by Anthropic in November, provides a standardized way to connect AI models to different data sources and tools, including data held by enterprises themselves. The major advantage of MCP is flexibility for AI users to switch between large language models (LLMs) and their vendors, according to Anthropic, developer of the Claude AI models.

This flexibility enables CIOs to choose between AI models based on what model delivers the best performance for the organization’s needs, says Jim Piazza, vice president of AI and predictive systems at managed services provider Ensono. It also helps them to avoid vendor lock-in, he adds.

“As models get more specialized, that’s where MCP has an opportunity for us to provide a little bit of order to the chaos,” he says. “I affectionately refer to MCP as the plumbing stack. It connects everything together.”

MCP also has a growing number of pre-built integrations that an LLM can plug into. In March, Microsoft announced MCP support in its Copilot Studio customization and agent-building tool. The integration allows Copilot Studio to add new AI apps and agents through MCP. Other AI vendors have also announced MCP compatibility in recent weeks.

MCP will also help DevOps teams take advantage of AI, some advocates say.

Explaining ACP

Earlier this year, following the release of MCP, IBM announced a draft of the Agent Communication Protocol, designed to enable AI agents, even those from different vendors, to connect to each other.

ACP is “a universal protocol that transforms the fragmented landscape of today’s AI agents into inter-connected teammates,” writes Sandi Besen, ecosystem lead and AI research engineer at IBM Research, in Towards Data Science. “This unlocks new levels of interoperability, reuse, and scale.”

ACP uses standard HTTP patterns for communication, making it easy to integrate into production, compared to JSON-RPC, which relies on more complex methods, Besen says. The protocol is part of an AI ecosystem, including BeeAI, that IBM donated to the Linux Foundation in April. 

Along comes Agent2Agent

Then, also in April, Google unveiled the competing Agent2Agent AI protocol, which also enables disparate AI agents to interoperate with one another. “Businesses benefit from a standardized method for managing their agents across diverse platforms and cloud environments,” Google developers wrote in a blog post. “We believe this universal interoperability is essential for fully realizing the potential of collaborative AI agents.”

Agent2Agent, supported by more than 50 Google technology partners, will allow IT leaders to string a series of AI agents together, making it easier to get the specialized functionality their organizations need, Ensono’s Piazza says.

Both ACP and Agent2Agent, with their focus on connecting AI agents, are complementary protocols to the model-centric MCP, their creators say.

With the emergence of the new protocols, Piazza can envision AI agent stores springing up, allowing users to pick from a menu of specialized agents or models from multiple vendors.

“Say there’s 100,000 models out there,” he says. “Why do I need to go train my own model, if I can get decent results out of a pre-existing, pre-hosted model, and I can just call it using MCP?”

Ultimately, the new protocols point to a new path to scalable AI adoption, says Christian Posta, global field CTO at cloud management vendor Solo.io.

“AI is already moving extremely fast, but speed without standardization just leads to chaos,” he says. “Standard protocols are the difference between going fast in a haphazard fragmented direction, repeating yourself, proliferating mistakes, and scaling with intention.”