A new book teaches developers to connect AI agents to tools with MCP and A2A

A shift from using to connecting

As aitimes.com reported in its AI Leader's Bookshelf column, developers who used to ask how to work with a coding agent are now asking how to hook an agent up to their company's internal tools. The column treats this as a change in the weight of the question: from use to connect.

The book it recommends is the Korean edition of a guide to MCP and A2A, published by Insight in August 2026. The author is Furukawa Hidekazu, a Japanese developer who has written many technical books, and the translator is credited as 두발타고.

The column also quotes Lee Won-jun, a technology merchandiser at Youngpoong Bookstore. He says interest is moving "from using tools to connecting them," and that more customers this year are asking for books on building MCP servers.

Two protocols, two jobs

A protocol is a shared set of rules that lets different software talk to each other. The book covers two of them, and their roles are clearly separate.

• MCP (Model Context Protocol) connects an agent to outside tools and databases.
• A2A (Agent2Agent) connects one agent to another so they can work together.

The author's framing, as the column describes it, is that these standards help turn AI from something that answers questions into a colleague that actually carries out work tasks.

How the book is laid out

It starts with explanation and then moves step by step through testing and building, with sample code along the way. The column gives a chapter-by-chapter picture.

Chapters 1 and 2 cover the architecture and server testing. In chapter 2, readers run public servers such as Playwright, Memory and DeepWiki one at a time to get a feel for how they behave.

Chapter 3 tries the same approach on seven hosts, meaning the apps or kits that run an agent: Cursor, GitHub Copilot, Claude Code, Gemini CLI, Codex CLI, the OpenAI Agents SDK and Google ADK. The point the column draws out is that the tools differ but the connection rules stay the same.

Chapter 4 has readers build servers for tools, resources, prompts and sampling, then write the client too. It ends by bundling a self-made server into a dxt file, a packaged form that makes it practical to picture rolling it out inside a company.

Chapter 5 turns to A2A, where agents exchange messages and tasks.

Who should start where

The bookstore suggests readers skip around depending on their situation.

• If MCP is a new term, read chapters 1 and 2 first.
• If you already use Cursor, Claude Code or Gemini CLI, go to the matching entry in chapter 3.
• If you are a back-end developer who must connect internal systems to an agent, chapter 4 is the core.
• If your team is planning several agents that cooperate, jump to chapter 5.

Kwon Ki-won, a team leader at the bookstore, describes the book as a light introductory guide that still makes readers build a server and a client by hand. In her view the length suits a developer who is new to agent connections and wants to follow it through over a weekend.

What to make of it

This is a book recommendation, not a product launch or a policy change, and the source gives no sales figures or reviews. It is a useful signal about where practical interest is heading, and it fits a wider pattern in today's news: other companies are also shipping MCP connections for their own products.

For non-engineers, the takeaway is simple. When you evaluate an agent tool, ask which standards it supports, because those decide what it can plug into. For builders, the book offers a structured way to learn by doing instead of piecing together scattered documentation. Note that the source describes the Korean edition, and says nothing about an English one.