Agentic Development with Vaadin

An AI coding agent writing a Vaadin application needs two things that a general-purpose model doesn’t have on its own.

The first is knowing the API. Vaadin adds components and methods faster than a model’s training data is refreshed, so an agent working from memory writes code against a version of Vaadin that may never have existed. The Vaadin MCP server answers from the real documentation for the version in your pom.xml.

When the application needs a third-party add-on, the separate Vaadin Directory MCP server lets the agent search the add-on catalog, retrieve installation instructions, and check Vaadin version compatibility. Despite the similar name, this isn’t the Vaadin MCP server: the Directory server searches add-ons, while the Vaadin MCP server searches documentation.

The second is knowing whether the code runs. Writing the code is the part agents are good at; finding out whether the change actually took effect is the part they’re bad at. Left alone, an agent rebuilds, relaunches the application, takes a screenshot, and reasons about a picture. The Dev Loop CLI replaces all of that with one command whose exit code is the verdict.

Both halves come with skills — instruction files that teach an agent when and how to use them. The skills are installed into your project and committed alongside it, so every agent and every developer on the repository works the same way.

Start with the Dev Loop Tutorial if you’d rather see the whole setup working before reading about the pieces.