MCP Dev Summit Seoul 2026

Two years ago, getting an AI agent to reach anything outside itself meant writing a whole bunch of glue code. A custom integration for every tool, every data source, every model you wanted to support. Nobody enjoyed this, and yet everybody did it anyway.

MCP was the emerging alternative - one protocol, implemented once on each side. It empowered all imaginable agents to be able to talk to anything through a standard set of conventions.

Two years later, the ecosystem around MCP is bigger than the spec itself. There are servers for nearly everything anyone would want to wire an agent into, clients across every major platform, and entire categories of capabilities that make the protocol suitable for more than just ferrying text-based data back and forth.

MCP is now the undeniable substrate most agentic workflows run on. Getting there took a couple of years, a large number of engineering debates, and the work of a community of maintainers, developers, and advocates that spans the globe. This is the story of us getting here.

Moving MCP tooling and agents from prototype to production creates a new set of challenges. Some are widely discussed protocol and infrastructure issues: scalability; agent communication; authentication. But as organizations work through those challenges, another bottleneck comes to the front: the operational governance layer around MCP adoption: trust, context, policy, provenance, accountability, and ecosystem participation

That layer is familiar territory for open source management. For years, Open Source Program Offices (OSPOs) practices have coordinated across legal, security, engineering, and product teams, helping decide which open source software to trust, how to contribute safely, how to manage provenance, how to define reusable policies, and more

As MCP makes it easier for agents to connect to tools, data, and internal systems, teams also need to ask: Which MCP servers are safe to use? Who maintains them? What context should agents read? What internal policies should be enforced? How do we trace what an agent did and under what rules?

Building on preliminary findings from the 2026 State of the OSPO survey, early stories from open source managers and strategists across organizations, and emerging agentic AI use cases being tested in the field, this keynote explores how OSPO practices can help organizations adopt MCP in production through ways that are trustworthy, auditable, context-aware, and connected to the AAIF ecosystem.

To attach an MCP server to each of hundreds of Naver Cloud Platform services, every service team had to implement MCP from scratch.
We built an MCP Generator to solve this — automatically generating MCP server code from OpenAPI specs.

Then Claude's MCP Builder Skills arrived. Suddenly, anyone could build an MCP server through a few conversations with Claude. The value of the generator we had built disappeared overnight.

We had to change direction.
And the question changed with it — from "how do we make building MCP servers easier?" to "how do we operate hundreds of them?"

That's how MCP Gateway was born. It aggregates multiple MCP servers behind a single HTTP endpoint, so users can connect every cloud service tool to their AI agent with just one URL and one API key.

This talk is structured in two parts:

Part 1 — The Journey: The process of figuring out how to provide good MCP servers for hundreds of cloud services — what problems we found, what we tried, and how the arrival of MCP Builder Skills became the turning point.

Part 2 — MCP Gateway: An introduction and demo of what problems it solves today, and a look at what we're building next.

In this workshop, you'll build a complete MCP-powered agentic system in Java from scratch.

We'll start by building an MCP server. You'll see how familiar the programming model is: annotate a method with @Tool, configure a transport, and your service speaks MCP. We'll connect it to a real-world API so the tools do something useful, not just "hello world."

Next, we'll build an AI agent that discovers and calls those MCP tools through natural language. You'll see how dependency injection and type safety make multi-tool composition clean and testable, and how hot reload lets you iterate on agent behavior in seconds.

Finally, we'll compose multiple agents into a workflow, including a supervisor agent that autonomously routes tasks to specialist agents, each with their own tools and MCP connections.

All materials are open source and based on a battle-tested workshop delivered to developer audiences worldwide. Attendees leave with running code and reusable patterns they can apply immediately.

Most people use Codex one prompt at a time. In this hands-on workshop, participants will turn Codex into a long-running research-to-build system by defining a clear Goal, providing rich project context, and steering the work as it progresses. We will assign Sol as the lead agent, delegate research and building tasks to Terra, use Luna for fast and repeatable work, then return to Sol for integration and evidence-based review. Participants will learn how to collaborate with Codex without taking implementation back, and how to verify results through tests, artifacts, and observable evidence.

Most people use Codex one prompt at a time. In this hands-on workshop, participants will turn Codex into a long-running research-to-build system by defining a clear Goal, providing rich project context, and steering the work as it progresses. We will assign Sol as the lead agent, delegate research and building tasks to Terra, use Luna for fast and repeatable work, then return to Sol for integration and evidence-based review. Participants will learn how to collaborate with Codex without taking implementation back, and how to verify results through tests, artifacts, and observable evidence.

Most people use Codex one prompt at a time. In this hands-on workshop, participants will turn Codex into a long-running research-to-build system by defining a clear Goal, providing rich project context, and steering the work as it progresses. We will assign Sol as the lead agent, delegate research and building tasks to Terra, use Luna for fast and repeatable work, then return to Sol for integration and evidence-based review. Participants will learn how to collaborate with Codex without taking implementation back, and how to verify results through tests, artifacts, and observable evidence.

When people talk about MCP, they think of a server protocol for desktop AI — Claude Desktop reading the filesystem, Cursor calling git, MCP servers humming on cloud infrastructure. But there's a whole other world that rarely shows up in the conversation: phones, smart watches, glasses, cars, robots, embedded edge devices. They all have mature SDKs, many can run small AI models locally, and almost none have an agentic tool layer to speak of. The natural question is: why isn't MCP there yet?
Turns out it can be — and is starting to be. This talk is about MCP in the mobile and broader smart-device world: what scenarios already work today, what's just landing, and what becomes possible when a device is no longer just a client to a cloud agent but a peer in a multi-device agentic system. We'll cover phones that host their own MCP servers exposing device capabilities, phones running an entire agent loop on-device with no network, system-level integrations Apple and Google are bringing through App Intents and AICore, and hybrid setups where local and cloud servers share one tool registry — then extend the picture beyond phones to glasses, cars, watches, and embedded edge