AI tooling for webforJ

Your AI assistant writes the Java. Now it can check it.

AI coding tools are fast, and their output still needs checking. webforJ gives them a loop of write, check and review: the real API to read, the right workflow to follow, and a test that returns pass or fail without a browser.

Install the plugin Step through the loop

For Claude Code, GitHub Copilot, Cursor, Gemini and Codex.

$claude plugin marketplace add webforj/webforj-ai

$claude plugin install webforj@webforj-ai

#Verify inside Claude Code

>/plugin

>/mcp

webforj appears under Installed

plugin:webforj:webforj-mcp appears under connected servers

The install commands from the docs. One install brings the MCP server and the Agent Skills.

Why a loop

Three ways agent-built Java breaks, and the check for each

When a coding agent has nothing to look things up in, it writes from memory. Telling it to be careful fixes none of these. Each failure needs its own check, and webforJ closes each gap with a piece of tooling.

01
Without a check
It calls an API that does not exist, or used to.
The real API
The MCP server searches the webforJ knowledge base, so the agent reads signatures instead of recalling them.
Read the docs
02
Without a check
It follows generic Java habits instead of the conventions of the framework.
The right workflow
Agent Skills tell the agent how to approach forms, security, styling, timers, and more.
Read the docs
03
Without a check
It produces code that compiles and does nothing useful when a user clicks.
A pass or fail
The TestBed runs a view in a plain JUnit test, so the agent can prove its own work.
Read the docs

The loop

Write, check, review

The agent writes the view, writes the test, runs it, and reads a pass or a fail. Step through one real screen, from the prompt to the passing test.

Your prompt

Add a webforJ greeting view at /greeting. A name field, a Greet button, and a paragraph that says Hello with the name.

A view is a Java class with a route

Nothing else is needed: no template, no JavaScript, and no REST layer between the screen and your code.

With the MCP server connected, the agent has looked up Composite, Button and @Route before it writes them.

GreetingView.java

@Route("greeting")
public class GreetingView extends Composite<Div> {
  private final TextField name = new TextField("Name");
  private final Paragraph message = new Paragraph();
  private final Button greet =
      new Button("Greet", ButtonTheme.PRIMARY, e -> greet());

  public GreetingView() {
    getBoundComponent().add(name, greet, message);
  }

  private void greet() {
    message.setText("Hello, " + name.getValue() + "!");
  }
}

Your prompt

Write a TestBed test for GreetingView that types a name, clicks Greet, and checks the paragraph.

The test follows the steps of a user

ComponentLocator finds the field by its label and the button by its text.
UserInteraction types and clicks, so the listener of the button really runs.
verifyThat checks the text a user would read.

GreetingViewTest.java

@WebforjTest
class GreetingViewTest {

  @Test
  void shouldGreetTheNameTheUserTyped(WebforjTestbed context) {
    context.add(new GreetingView());

    var name = ComponentLocator.of(TextField.class).byLabel("Name").locate();
    var greet = ComponentLocator.of(Button.class).byText("Greet").locate();

    UserInteraction.type(name, "Ada");
    UserInteraction.click(greet);

    verifyThat(ComponentLocator.of(Paragraph.class).locate())
        .hasText("Hello, Ada!");
  }
}

The agent reads a pass or a fail

The agent runs mvn test and reads the result. No browser is opened and no server is started.

If it fails, the message says what differed, and the agent edits and runs again. A failure it can read is a failure it can fix.

You review a result that already passed instead of a guess.

Terminal

$mvn test

GreetingViewTestshouldGreetTheNameTheUserTypedPassed

When a check fails, the message reads like thisExpected TextField to have value <Grace> but it has <Ada>.It names the type of the component, the expected value and the actual one.

Stage 1 of 3

The pieces

01 webforJ AI plugin

One install brings both pieces

The plugin bundles the MCP server and the Agent Skills. Together they turn an assistant that guesses at webforJ conventions into one that follows them.

Mention webforJ in your prompts. That is the cue that sends the assistant to the plugin instead of general Java knowledge.

Install the plugin

Where it installs

Claude CodeMarketplace command
GitHub Copilot CLIMarketplace command
VS Code + CopilotChat: Install Plugin From Source
Gemini CLIExtension install
OpenAI Codex CLIMarketplace command, then /plugins
Cursor, Kiro, Goose, Junie, AntigravityManual configuration

What a prompt reaches for

Wrap this Custom Element library as a webforJ component.creating-components skill
Style this view with the DWC design tokens.styling-apps skill
Scaffold a new webforJ sidemenu project called CustomerPortal.MCP project scaffolder
Generate a theme from brand color #6366f1.MCP theme generator
Find the webforJ docs on @Route and routing.MCP knowledge search

01a MCP server, part of the AI plugin

The assistant reads the real API instead of recalling it

Model Context Protocol is an open standard that lets AI assistants call external tools on demand. The webforJ MCP server answers from the real framework: docs, design tokens, component styling and project scaffolding.

Validation catches typos and invented tokens before they ship as silently failing CSS.

See what the server can do

Target the right webforJ version
Reads pom.xml when available and otherwise asks you. Every answer after that is scoped to your version.
Look things up in the knowledge base
An API question, a guide, a code sample or the Kotlin DSL, answered from the real framework.
How do I set up routing with @Route in webforJ?
Scaffold a new project
The correct Maven archetype command from hello-world, blank, tabs or sidemenu, in the webforj or webforj-spring flavor.
Create a webforJ project called CustomerPortal using the sidemenu archetype
Generate a DWC theme
Primary, success, warning, danger, info, default and gray palettes from one brand color, with automatic text contrast.
Generate a webforJ theme from brand color #6366f1
Style DWC components correctly
CSS custom properties, shadow parts, reflected attributes and slots, read before any CSS is written.
What CSS variables and parts does dwc-button expose?
Validate design tokens
Checks CSS, Java or Markdown against the real --dwc-* catalog and flags unknown names with suggested corrections.
Validate app.css for any unknown --dwc-* tokens

Only the MCP server, in Claude Code

claude mcp add webforj-mcp https://mcp.webforj.com/mcp -t http -s user

The plugin is the recommended install. This command connects the server alone, at https://mcp.webforj.com/mcp.

01b Agent Skills, part of the AI plugin

The right workflow for the task in front of it

The MCP server makes accurate information available. A skill tells the assistant when to look something up, which approach fits, and what order to do things in.

Skills follow the open Agent Skills standard. The assistant loads one automatically when your prompt matches it.

Browse the skills

webforj-adding-servlets
REST endpoints, webhooks and custom servlets next to the webforJ UI.
Add a REST endpoint at /api/orders.
webforj-building-forms
Binding, Jakarta validation, masked inputs and responsive form layouts.
Build a registration form bound to my User bean.
webforj-creating-components
Reusable Java components from Custom Elements, JavaScript libraries or compositions.
Compose these webforJ components into a reusable card.
webforj-handling-timers-and-async
Interval, Debouncer, Environment.runLater and PendingResult instead of raw timers and threads.
Refresh this dashboard every 30 seconds.
webforj-localizing-apps
Message bundles, runtime language switching and browser locale detection.
Add multi-language support with English and Spanish.
webforj-securing-apps
Login, logout, role-based access and secure-by-default routes.
Protect /admin so only users with the ADMIN role can see it.
webforj-styling-apps
Themes, palettes and component styling with every --dwc-* token validated.
Theme this app with a blue palette.
webforj-upgrading-versions
Major version upgrades with the official webforj-rewrite OpenRewrite recipe.
Upgrade this app from webforJ 25 to 26.

02 TestBed

A clear pass or fail, with no browser and no server

A view runs inside a plain JUnit test. Components attach, fire their events, and navigate exactly as they do in a running app.

Every interaction finishes before the next line of the test runs, so nothing waits and nothing flakes. Spring Boot apps use the webforj-spring-testbed module.

Read the TestBed guide

FindComponentLocator
Describes a component the way a user picks it out, by its type and what it shows.
ActUserInteraction
Makes only the gestures a user could make. Clicking a disabled button fails the test.
CheckverifyThat
Checks what the user would see, and a failure names the component and both values.

pom.xml

<dependency>
  <groupId>com.webforj</groupId>
  <artifactId>webforj-testbed-junit</artifactId>
  <scope>test</scope>
</dependency>

The version comes from the webforJ BOM. A project created from an archetype already imports it.

03 craftforJ Assistant

A coding agent inside the running app

It writes Java freely, compiles each edit before you see it, and applies it to the running app. Applying a change restarts the app, and the agent picks its plan up again once it is back.

It ships with webforJ, so there is nothing to install. craftforJ ships no model of its own: add an API key for a supported provider or point it at a model running locally with Ollama.

Meet the assistant

Lives in: AI pluginYour editor; craftforJ AssistantThe running app
Reads: AI pluginYour source files; craftforJ AssistantYour app, live, with its real values
Verifies by: AI pluginYour next build; craftforJ AssistantCompiling each edit before you see it
Suited to: AI pluginBuilding something new from scratch; craftforJ AssistantUnderstanding, fixing and prototyping against the app in front of you

The craftforJ Assistant asking before it acts, inline in the transcript, with Apply, Allow for chat and Dismiss. — The assistant asking before it acts, inline in the transcript.

webforj.conf

webforj.debug = true
webforj.devtools.craftforj.enabled = true

craftforJ runs only in development. Open it with Alt + Shift + D and switch to the AI Assistant tab.

How to start

Ask for one screen with a test

Pick a screen you would otherwise build by hand this week. webforJ is the full-stack Java framework for business web apps: you build the whole app, screens included, in Java.

Install the plugin

claude plugin marketplace add webforj/webforj-ai
claude plugin install webforj@webforj-ai

2
Scaffold or open a webforJ project
Ask the MCP server for a project, or open one you already have.
3
Ask for the view and its TestBed test in one prompt
Describe the behavior you care about, so the test asserts it.
4
Run mvn test and read the result

Three ways agent-built Java breaks, and the check for each

Write, check, review

A view is a Java class with a route

Three pieces behind the loop

One install brings both pieces

Where it installs

What a prompt reaches for

The assistant reads the real API instead of recalling it

Target the right webforJ version

Look things up in the knowledge base

Scaffold a new project

Generate a DWC theme

Style DWC components correctly

Validate design tokens

The right workflow for the task in front of it

A clear pass or fail, with no browser and no server

A coding agent inside the running app

Ask for one screen with a test

Install the plugin

Scaffold or open a webforJ project

Ask for the view and its TestBed test in one prompt

Run mvn test and read the result

Run `mvn test` and read the result