AI Agent | MCP | Communication Protocol

The road to becoming an AI creator | Article introduction: The possibilities of AI agents are endless! Why will MCP and communication protocols accelerate the future of work and technological innovation? #AIagent #MCP #Communicationprotocol

Table of Contents

Video explanation

The future opened up by AI agents: A simple explanation of Model Context Protocol (MCP) and communication mechanisms!

Hello, I'm John, a veteran blogger. The world of AI has been evolving at an incredible speed recently! In particular, you may have heard the term "AI agent" more and more often. However, I'm sure there are many people who are wondering, "What is an AI agent?" or "It seems kind of difficult..." So, this time I'd like to talk about AI agents and the essential technology that makes them smarter and more convenient to use, "Model Context Protocol (MCP)"Or"Agent Communication Protocol" I will explain in simple terms, avoiding technical terms as much as possible, so that even beginners can understand. Let's take a look at the future of AI technology together!

What is an AI Agent? The Basics

First, let's start with the basics: what is an "AI agent"?

Conventional AI, for example chatbots like ChatGPT, mainly work by asking us questions and giving us an answer. It may be easier to understand if you think of this as a "Q&A mode."

on the other hand,AI agent (AI program that makes autonomous decisions and executes tasks) is an AI that is more proactive and can perform complex tasks. It is not just about answering questions, but when given a goal, it will try to achieve that goal.Plan your own actions, gather information, use the tools you need, and sometimes collaborate with other AI agents to complete multi-step tasks.Imagine it as a smart assistant that works on our behalf.

For example, say you ask an AI agent to "give me a third quarter revenue forecast for our cloud product." The AI agent will understand this instruction and autonomously create a plan to achieve that goal. How it does this will depend on the tools it has access to (e.g., tools to connect to internal databases, analyze data, generate reports, etc.).

The AI agent would first check what tools it has at its disposal, and then might proceed with the task through the following steps:

Controller (the parent agent that receives instructions and manages the overall flow) calls the planning agent.
The planning agent returns an execution plan.
The assessor agent reviews the execution plan.
The controller executes the plan using the Generate Database Query tool (GenSQL) and the Execute Database Query tool (ExecSQL).
Finally, the evaluation agent reviews the final results and advises revision and re-execution of the plan if necessary.

In this way, an AI agent can handle complex tasks by working with multiple agents and tools internally in response to a single instruction. This is the main feature of an AI agent, and it is the problem it is trying to solve, i.e. "Automating Complex Tasks" is the answer.

Unique features include:Autonomy (the ability to think and act independently),Planning ability (ability to think through procedures),Tool use (ability to use external tools)AndCooperativeness (ability to cooperate with other agents) .

Why are communication protocols important? Challenges for AI agents

In fact, there is a big challenge for an AI agent to work this intelligently.Interoperability (the ability of different systems and AI to work well together)" is a lack of.

Currently, the development of AI agents is said to be in the "pre-standardization stage," meaning that each company develops their own AI agents in their own way, making it difficult for AI agents made by different companies to smoothly exchange information or cooperate with each other.

Think about it. In our work, we need to access many data systems on a daily basis, such as customer management systems like Salesforce, internal wikis, and other business systems. If these systems were not connected or the data formats were different, work efficiency would drop dramatically. The same goes for AI agents. Without a standardized means of communication, it would just create a new type of "data silo" (a state in which information is isolated).

So what is important is,communication protocol - a set of common rules or procedures for communicating between computers" This is a set of rules that allow AI agents to communicate with each other, and between AI agents and external tools, in a "common language." This protocol allows different AI agents to work together smoothly and perform more complex and valuable tasks.

Let's take a look at the hot protocols! MCP, A2A, ACP

Currently, several protocols have been gaining attention for standardizing communication between AI agents. Each protocol focuses on solving slightly different problems. Here, we introduce three particularly important protocols: Model Context Protocol (MCP), Agent-to-Agent Communication Protocol (A3A), and Agent Communication Protocol (ACP).

What is the Model Context Protocol (MCP)?

First of all, we would like to introduceModel Context Protocol (MCP) developed by Anthropic (the company that developed Claude AI, a rival to ChatGPT)The MCP is a set of tools that allow AI agents and models to seamlessly integrate tasks, tools, and multi-step reasoning.Standardize how to manage, share, and use "context" (meaning context, situation, and related information that is important for AI to understand and execute tasks)The purpose is to:A bridge to connect AI agents to external tools and data sources.

MCP isClient-server architecture – a system consisting of a client that requests a service and a server that provides the service. In this approach, AI applications act as clients, requesting information from a server that provides access to external resources.

For example, all the dataApache Kafka – a distributed messaging system for processing large amounts of data in real time Let's say that all the data is stored in topics (where each type of data is stored). In this case, you can build a dedicated Kafka MCP server and run Claude, Anthropic's AI model, as an MCP client. The user instructs Claude to "connect to a Kafka broker (the core server of the Kafka system) and list all the topics." Thanks to MCP, Claude does not need to know how to access the Kafka broker specifically. Behind the scenes, Claude (the client) sends a request to the MCP server, which interprets the request and executes the relevant Kafka function.

To achieve this, we need to define what the MCP server can do. For example,handler.goThis file contains a list of functions that the server can handle (e.g., the ability to create a new Kafka topic). In this way, the MCP server exposes the capabilities of external systems, such as databases, to AI agents through a standardized interface.

The great thing about MCP is that it doesn't just give you the tools to use it.Prompt templates (standardized instructions) The advantage is that you can define what is considered confidential and put guardrails in place to protect sensitive or private information. This makes it easier to get accurate results while still protecting the privacy of patient data when working with medical databases, for example.

MCP is an open standard (a public standard that is freely available to anyone) and many companies, including Microsoft Azure and Cloudflare, are promoting its adoption.

What is Agent2Agent Protocol (A2A)?

Next, we will introduce"Agent-to-Agent Communication Protocol (A2A)" developed by GoogleThe protocol allows AI agents to communicate across multiple platforms without being tied to a specific framework or vendor.Allowing them to directly communicate with each other, collaborate, and work together to solve complex tasksThat is,Standardize collaboration between AI agents created by different developersThing.

A2A is characterized by "opaque communication," meaning that cooperating agents can exchange information without exposing or coordinating each other's internal structure or logic. This allows different teams and organizations to freely build and connect agents without adding new constraints.

Specifically, A2A requires that agents are described by metadata (data about data) in an ID file known as an "agent card". A2A clients send requests as structured messages to the A2A server and receive real-time updates for long-running tasks.

A useful example of A2A is the healthcare use case, where an agent in one healthcare organization communicates with an agent in another healthcare organization in a different region using the A2A protocol, where the agents need to ensure data encryption, authorization (e.g. OAuth/JWT), and asynchronous transfer of structured healthcare data using Kafka.

What is the Agent Communication Protocol (ACP)?

Finally,"Agent Communication Protocol (ACP)" developed by IBMIt is an open protocol for communication between AI agents, applications, and humans. According to IBM, "In ACP, agents are software services that communicate primarily through multimodal messages driven by natural language." The protocol is agnostic to how agents work internally and specifies only the minimum prerequisites necessary for smooth interoperability.

ACP and A2A are similar in that they eliminate agent vendor lock-in, accelerate development, and make community-built agents easily discoverable regardless of implementation details. However, there is one key difference:ACP leverages IBM's BeeAI open source framework to enable communication between agentsWhereas A2A helps agents from different frameworks communicate with each other, ACP is specifically focused on agent collaboration within IBM's BeeAI ecosystem.

The BeeAI project currently consists of three main core components:

BeeAI Platform – for discovering, running and configuring AI agents
BeeAI Framework – for building agents in Python or TypeScript
Agent Communication Protocol (ACP) – for communication between agents

Technical details: Learn more

Let's dig a little deeper into how these protocols specifically enable AI agents to communicate.

How MCP works
MCP acts as an "interpreter and adapter" that allows AI agents to "talk" to external tools and databases. When an AI agent sends a request in a standardized format to the MCP server, saying "I want this data" or "I want you to do this process," the MCP server understands it, translates it into words that the target tool or database can understand, issues instructions, and returns the results to the AI agent in a standardized format. This "context sharing" is important, as the AI agent can receive the information it needs to execute a task through MCP, just like a "cheat sheet." This allows AI agents to use a wide variety of external resources in a unified way.

How A2A and ACP work
A2A and ACP are rulebooks for AI agents to communicate directly with each other. They define common message formats and procedures that AI agents made by different companies can recognize each other (using metadata such as "agent cards") and understand. This allows AI agents with different specialties to team up to tackle bigger problems. For example, an agent good at research, an agent good at analysis, and an agent good at report writing can work together through A2A to complete a single report. ACP has a similar purpose, but is designed to be particularly compatible with IBM's BeeAI framework.

These protocols can be seen as fundamental technologies that will not only enable AI agents to become smarter on their own, but also build more advanced intelligent systems for society as a whole.

Development team and community: reliability and activity

Who are developing these leading protocols and how active is the community?

MCP (Model Context Protocol) :
The developer is known for its high-performance AI model "Claude."AnthropicAnthropic is also committed to AI ethics and has been highly praised for its technological leadership. MCP is published as an open standard, and its adoption is progressing in major cloud platforms, such as Microsoft Azure, which has made MCP support generally available. Companies such as Cloudflare also provide information about MCP, and the ecosystem is expected to expand.
A2A (Agent to Agent Communication Protocol) :
The developer is a well-known technology giant.GoogleGoogle is a world leader in AI research and development, with enormous technical capabilities and resources. A2A is also published as open source on GitHub, providing an environment where a wide range of developers can access and contribute.
ACP (Agent Communication Protocol) :
The developer has been providing enterprise IT solutions for many years.IBMIBM is also actively investing in the AI field, with particular strength in business applications. ACP is closely related to IBM's BeeAI framework and is being developed as open source.

Although these protocols are relatively new, they are promoted by companies with great influence in the AI field, and have strong backing. Communities are also gradually forming, and their future development and spread will be of great interest.

Use cases and future prospects: What can AI agents do?

So, in what specific situations will AI agents that utilize these communication protocols be useful, and how will they change our future?

Common AI agent use cases:

Automate complex business processesAI agents will take over routine tasks and information gathering and organization that previously took a lot of time from humans, such as market research, report writing, data analysis, and responding to customer inquiries.
Advanced Personal Assistant: It autonomously handles tasks such as schedule management, travel arrangements, and integration with multiple online services by understanding the user's intentions.
Software Development Support: It dramatically improves developer productivity by automatically generating code, running tests, finding bugs and suggesting fixes, and more.
Cybersecurity Automation：AI agents support security operations (SecOps) such as threat detection, analysis, and response, building a faster and more effective defense system. Swimlane and others have introduced this as an example of how MCP is used.

Values brought by the protocol:

MCPAI agents can now access a wider variety of tools and data sources, making them much more capable -- for example, an AI agent could use MCP to access real-time stock data to perform market analysis, or an AI agent could query Kafka topics in natural language.
A2A/ACP: It will be possible to build a "multi-agent system" in which multiple AI agents with different specialties cooperate to share and execute more complex tasks. For example, we can envision a future in which an AI agent specializing in information gathering, an AI agent specializing in data analysis, and an AI agent specializing in creating summary reports work together through A2A or ACP to automatically generate a single, high-quality research report for a certain research topic. Safe data sharing between different institutions in the medical field is also expected.

Future Outlook:
With the development of these technologies, AI will become more advanced and the scope of its autonomous operation will expand. AI will be seamlessly integrated into various applications and industries, potentially creating new services and value that were previously unimaginable. AI agents are expected to play an active role as "practical, sustainable solutions to complex real-world problems."

Comparison with competitors: strengths of each

What are the strengths of MCP, A2A, and ACP? Rather than being direct competitors, they are protocols that have been created to address different challenges faced by AI agents and have complementary aspects.

MCP (Model Context Protocol) :
Strengths: It specializes in linking AI agents with external tools and data sources, and in "context management." It is a very powerful foundation for individual AI agents to access external resources and efficiently obtain the information they need to execute tasks. It has a clear purpose: "connecting agents to tools and data."
A2A (Agent to Agent Communication Protocol) :
Strengths: The aim is to standardize direct communication between agents without relying on a specific framework. It is effective when building an open multi-agent system in which AI agents created by various developers work together. The main focus is on "standardizing collaboration between agents."
ACP (Agent Communication Protocol) :
Strengths: It is optimized for agent-to-agent communication within IBM's BeeAI ecosystem. Developers who are already using or considering using the BeeAI framework can expect seamless integration. It is "focused on BeeAI agent collaboration."

Currently, Anthropic's MCP has attracted attention first, and in response, Google released A2A and IBM released ACP. In the future, we may see scenarios where these protocols are used in combination.

Risks and Cautions: What you need to know

AI agents and these communication protocols have great potential, but there are also some risks and precautions to be aware of when using them.

Technology MaturityNote: These protocols are still in development and are in the "pre-standards" stage, so it is important to take into account that specifications may change or new standards may emerge in the future.
Complexity: Building high-performance AI agents and managing the systems in which they communicate smoothly can still be a complex task.
Security: As AI agents gain access to more tools and data, security becomes a top priority. Robust security measures are essential to deal with new threats, such as data leakage due to unauthorized access and system failure due to unintended operations. Security companies such as Zenity.io and Orca.security have also sounded the alarm about security risks associated with AI agents and MCPs.
Reliability and UnpredictabilityThe behavior of AI agents is not always 100% predictable, and they may produce unexpected results, especially when faced with complex tasks or unknown situations.
Ethical concernsAs AI agents that can make autonomous decisions and act autonomously are used in various aspects of society, it will become necessary to discuss ethical issues such as the criteria for making decisions and where responsibility lies.

Fully understanding these risks and careful design and operation are essential for the healthy development of AI agent technology.

Expert opinions and analysis

What do experts in the field see as the current and future of AI agents and communication protocols?

Adi Polak of Confluent (a data streaming technology company) points out that "AI agents are in a 'pre-standards stage'. Without standardized communication for AI agents, we're just building a new type of data silo," and his colleague Sean Falconer adds, "Today's AI agents lack the interoperability they need to not just do something, but actually do meaningful work."

These comments suggest how important it is for communication protocols like MCP, A2A, and ACP to become standardized and adopted for AI agents to be truly valuable. There is active discussion in the industry about which protocols will become the standard and how they will coexist and evolve.

IBM explains its design philosophy for ACP as "independent of how agents work internally, specifying only the minimum prerequisites necessary for smooth interoperability," which shows that the company values openness. Anthropic's MCP documentation is also an important source of information for understanding their vision.

Overall, experts recognize these protocols as key to accelerating the evolution and practical application of AI agents, and are excited about their development.

Latest News and Roadmap Highlights

The field of AI agents and communication protocols is evolving at a very fast pace. Let's take a look at some of the latest developments.

Emergence of major protocolsAfter Anthropic's MCP attracted attention, Google released A2A and IBM released ACP in quick succession. This shows that the movement toward standardizing collaboration between AI agents and between agents and tools is gaining momentum.
MCP Support on Microsoft Azure: Microsoft has made "Agent Mode" for building AI agents using MCP (Model Context Protocol) generally available on Azure. This is big news as it shows that MCP is beginning to play an important role in practical application development.
Activating the open source community: Developer communities are actively discussing and contributing to each protocol through their GitHub repositories. New ideas for tool integration and proposals for protocol improvements are being generated every day.
Future roadmap: We will have to wait for the details from each developer to announce the specific roadmap, but in general, the focus is expected to be on further dissemination of the protocol, expansion of compatible tools (especially MCP), improved ease of use, and strengthening of security functions. In addition, research and development will progress toward cooperation between different protocols and the realization of more advanced multi-agent systems.

This field is evolving rapidly, so we encourage you to keep an eye on announcements from major players such as Anthropic, Google, and IBM, as well as related tech news. Prioritizing your learning and adoption is key, and adopting protocols, tools, and approaches that save you time and effort will be key to succeeding in the fast-changing world of AI.

よくある質問 (FAQ)

Q1: How is an AI agent different from previous AI?: A1: Conventional AI is often specialized for a specific task, such as a chatbot that returns one answer to one question. On the other hand, an AI agent is given a goal and can think of ways to achieve it.It can autonomously execute complex, multi-step tasks by making plans, gathering information, using the necessary tools, and sometimes working with other AI agents.It's like having a smart assistant that can think and act on its own.
Q2: What is MCP (Model Context Protocol) for?: A2: MCP is a system that allows AI agents to communicate with external databases and APIs (a mechanism for connecting software).Standard "procedures" for efficiently connecting to tools and data sources and sharing the necessary information (context) accuratelyThis allows AI agents to use a variety of information sources to become smarter and more adept at their tasks.
Q3: There are so many protocols, which one should I use?: A3: The protocol you choose depends on what you want your AI agent to do.MCPis great at connecting AI agents with tools and data.A2Ais designed to allow AI agents created by different developers to communicate with each other.ACPIn particular, we focus on agent-to-agent communication in IBM's BeeAI development environment. Each of these aims to solve slightly different problems, so they may be used in combination in some cases.
Q4: Is it safe to use AI agents?: A4: As AI agents get smarter and have access to more data and tools,Security is a very important issueIn order to prevent unauthorized access and misuse of data, it is essential to choose a secure protocol and design the system with security in mind. Both developers and users need to be highly security conscious.
Q5: Are these technologies already in practical use?: A5: Yes, some of them have already been put to practical use. In particular, MCP is being supported by major cloud services such as Microsoft Azure and is beginning to be used in specific applications. However, overall, it is still a new technology field,Currently developing rapidlyIt is expected that many more practical applications will emerge in the future, bringing about major changes to our lives and work.

A new era of AI agents: MCP and communication protocols will change the future