Blog

Your Vaadin app as a Spring AI MCP server and client

By  
Olli Tietäväinen
Olli Tietäväinen
·
On Nov 19, 2026, 2:15:16 PM
·

If you've used an agentic development tool for your Vaadin application's development, you're probably familiar with the Vaadin MCP server, which gives your coding agent a direct link to the latest Vaadin documentation. But documentation helpers for coding assistants are only one use for MCP. An MCP server can not only provide answers to questions; it can provide tools for agents to invoke any kind of operation your server is capable of. So your Vaadin app could also be an MCP server, giving desktop assistants, IDE agents and chat tools a way into its business logic and data. That's what the new one-day MCP and Vaadin applications training is about.

If your Vaadin app already exposes a REST API for external clients, you've already had a similar idea: an MCP server just opens an API for agentic callers instead of REST clients. But don't just wrap every REST endpoint in an MCP tool and call it a day. REST endpoints are traditionally built to be composable, data-heavy and versioned. For agentic callers, you'll want human-readable tool descriptions that match real-life tasks, because the agent starts from what its user asks. I'm not going to write "Find the ORDER_AMOUNT from ORDERS where CUSTOMERID equals my customer id and sum the ORDER_AMOUNT and print true if the sum is greater than or equal to the free shipping threshold" into a chat box, ever. I could write "Am I eligible for free shipping yet", and your MCP server could provide a tool that can answer this specific question.

That previous example was about a read-only tool, but there's no reason an MCP server can't offer tools that write or delete data. At this point, if not earlier, you should start being concerned about security, though. Authorization is optional in MCP: the specification defines an OAuth 2.1-based flow for servers that run over HTTP, but a server that doesn't implement it is open to all the clients that can see it. It's again not that different from a REST API: it's up to the server to authenticate and authorize all callers.

Your Vaadin app can be an MCP client, too: if it has an AI-powered chat, you can make the chat more powerful by giving its LLM integration access to MCP tools. The protocol even offers lesser-known features such as Elicitation, where the server asks the user, through the client, for confirmation or more information. In a Vaadin app, that can be a dialog in your UI.

The MCP and Vaadin applications training covers all of this in one day. It opens with a part on what MCP is and isn't, with no code, followed by three hands-on parts: connecting a Vaadin chat view to an MCP server, exposing your own app as an MCP server, and securing and testing that server. It's meant as a standalone follow-up to the Building AI-powered applications with Vaadin training, where we cover the basics of LLM integrations, AI chats in Vaadin, tool calling, and RAG. If you already know the Spring AI basics, you can start with this one. The foundation for both trainings is Vaadin 25 and Spring AI 2.

The training runs live, online or onsite. Onsite is by agreement, with travel on top. The price is per session, not per participant, excl. VAT: 950€ for this one-day training, or 1,800€ for two days when you add Building AI-powered applications with Vaadin. Read about MCP and Vaadin applications and register your interest.

Olli Tietäväinen
Olli Tietäväinen
Olli is a senior software engineer who has been working at Vaadin since 2015.
Other posts by Olli Tietäväinen