AI Creator's Path News: Introducing three protocols that pave the way for the future of AI agent development: MCP, A3A, and ACP! #AIAgent #MCP #A2A
Video explanation
Will AI "little secretaries" get smarter? What is the future AI technology "protocol"?
Hello to everyone interested in the world of AI and technology! I'm John, an AI blogger. You may have heard the term "AI agent" more often recently. You may think, "It's something like ChatGPT, right?" But it's actually a more advanced AI technology that sounds like something from a science fiction movie.
AI agents can not only answer questions, but also plan their own tasks and work with various tools and data to complete complex tasks that span multiple steps. For example, in the future, AI agents may be able to easily do requests such as "make a sales forecast for next month and report it to the relevant departments by email."
However, there is one big challenge to overcome before these intelligent AI agents can truly take advantage of the power of the technology."How can AI agents talk to each other, or to other systems, smoothly?"Today, I will explain the "conversation rules," or "protocols," of these AI agents in a way that is easy to understand even for beginners!
Why do AI agents need "conversation rules"?
Imagine if you went abroad and tried to accomplish a big project with people who didn't speak your language at all... It would be difficult, right? You wouldn't know what each other wanted to do or how to cooperate, and the work wouldn't progress very quickly.
The same is true in the world of AI agents. Even if there are AI agents with various capabilities, they cannot work together unless there is a common rule (protocol) for them to communicate with each other. Each AI agent becomes like an "isolated island" and cannot fully utilize its capabilities. In technical terms, this is called "data siloing (a state in which information is fragmented and cannot be utilized)."
That is why, in order for AI agents to work more conveniently and efficiently, we need a common language and rules that everyone can understand.Standardized ProtocolsIt becomes very important.
What exactly does an AI agent do?
Before we get into protocols, let's look at a simple example of how an AI agent actually works.
For example, if you ask an AI agent to "give me a third quarter revenue forecast for our cloud product," the AI agent (let's call it the "controller" here, as it acts like a command center) will act like this:
- First, check what you can do (what tools you have). For example, you might have a "tool to make a plan," a "tool to get information from a database," and a "tool to do a final check."
- The controller instructs the "planner" AI agent to "make a plan to generate revenue forecasts!"
- The plan created by the planning agent is reviewed by the "judging" AI agent, who asks "Is this plan okay?"
- Based on the approved plan, the controller instructs the "database operation" AI agents to "create this SQL (a command statement for operating the database) and execute it!"
- The final result will be checked by the assessment agent again and reported to you if there are no problems. If any corrections are required, we will start over from the beginning of the plan.
In this way, many AI agents and tools work together behind the scenes to complete a single task. The protocol that I will talk about next is what makes this collaboration smooth.
Top 3 AI "chat rules" to watch! What are MCP, A2A, and ACP?
There is currently a lot of activity in standardizing communication between AI agents, and several promising protocols have emerged. Here are three that are particularly noteworthy!
1. MCP (Model Context Protocol) – A translator connecting AI and tools
MCP is a protocol developed by Anthropic (a company famous for its AI called Claude).Common rules for AI agents and models to communicate effectively with various tools and data sourcesThis defines the following.
To put it in more complicated terms, we use a system called "client-server architecture." This means that the AI application (the client, i.e. the customer) requests information and processing from the server (the store clerk). Even if the AI agent does not know the details of how to use the data storage location (for example, the data storage system called Apache Kafka), it can simply ask the MCP server to "get that data," and the server will carry out the appropriate processing on its behalf.
For example, if you want to see a list of data in Kafka, by using MCP, an AI agent can simply ask the MCP server to show them a list of topics (data containers) without having to have specialized knowledge of Kafka. It's very convenient!
2. A2A (Agent2Agent) Protocol – AIs talk directly to each other!
A2A is a protocol developed by Google, and as the name suggests,AI agents communicate directly with each other and work together to solve complex tasksIts unique feature is that it allows various AI agents to work together without being tied to a specific company's system.
In A2A, each AI agent has a self-introduction file called an "agent card." This makes it easier for other AI agents to understand what the agent can do.
For example, let's say an AI agent in one hospital needs to securely send patient information to an AI agent in another hospital in another region. By using the A2A protocol, it is possible to exchange information safely and reliably, encrypting data and providing authentication (like identity verification), even without knowing the detailed internal workings of each other's systems.
3. ACP (Agent Communication Protocol) – Smooth integration in specific environments
ACP is a protocol developed by IBM that aims to facilitate communication between AI agents, applications, and humans. Similar to A2A, ACP aims to allow AI agents to communicate with each other without being tied to a specific vendor.
However, there is one big difference. ACP places special emphasis on efficient communication between AI agents developed with IBM's "BeeAI" open source framework (a foundation for easy development). In other words, it is designed to enable AI agents to cooperate more smoothly within the BeeAI ecosystem.
So what is the difference between these protocols?
We've looked at three protocols so far, each of which has slightly different areas of expertise. To summarize briefly, they are as follows:
- MCP (Anthropic): Various AI agentsTools and DataRules for chatting well.
- A2A (Google): AI agents from different companiesHowever, these are rules that allow people to cooperate directly without knowing each other's detailed mechanisms.
- ACP (IBM): Mainly IBMAI agents built within the BeeAI frameworkHowever, these are rules to ensure smooth cooperation.
It's not a matter of which protocol is better or worse, but rather that they are used according to their respective purposes, and in the future, they may be used in combination.
The future of AI agents and protocols
The world of AI agents is still in its infancy. These protocols are currently under active development, with Google and IBM announcing A2A and ACP reportedly following the success of Anthropic's MCP.
The development of these "conversation rules" is essential for AI agents to become smarter and more useful in our daily lives and work. Standardizing the rules will enable developers to create AI agents more easily and efficiently, and we can also expect to see the emergence of more convenient and easy-to-use AI services for users.
Personally, I can't help but get excited when I imagine a future where AI agents automatically connect with various systems and do tedious tasks on our behalf! For example, when planning a trip, all you need to do is ask an AI agent to "find and book a good lodging and transportation with a budget of 3 yen" and it will make all the arrangements for you.
To that end, I hope that protocols like the ones we introduced here will continue to evolve, allowing AI agents to "chat" more freely and more intelligently!
This article is based on the following original articles and is summarized from the author's perspective:
A developer's guide to AI protocols: MCP, A2A, and
ACP