Browser automation for Kotlin Multiplatform.
Write a workflow once and run it in Android WebView, iOS WKWebView or JVM Chromium.

Website · Documentation · Examples

A Playwright-inspired Kotlin API for locators, assertions, networking and screenshots, plus compact commands an AI agent can drive the same browser with.

[!NOTE] 0.0.2 alpha. Available on Maven Central. APIs and platform coverage may change.

One dependency in commonMain covers Android, iOS and JVM. Gradle picks the platform variant.

kotlin {
    sourceSets {
        commonMain.dependencies {
            implementation("io.github.bernaferrari:kmpwright-core:0.0.2")
        }
    }
}

Requires mavenCentral(). For the Android Compose wrapper, add io.github.bernaferrari:kmpwright-compose:0.0.2 to androidMain.

Fill a form, click its button and check the result:

import com.bernaferrari.kmpwright.core.*

suspend fun demo(page: Page) = with(page) {
    setContent(demoHtml)
    getByLabel("Name").fill("Ada")
    getByRole("button", name = "Say hello").click()
    expect(getByRole("status")).toHaveText("Hello, Ada!")
}

The page shows Hello, Ada! and the assertion passes. Locators wait until an element can be used; assertions wait for the expected result. No server or account needed.

val demoHtml = """
    <!doctype html>
    <html lang="en">
    <meta charset="utf-8">
    <title>KMPwright demo</title>
    <label for="name">Name</label>
    <input id="name">
    <button onclick="document.getElementById('greeting').textContent =
        'Hello, ' + document.getElementById('name').value + '!'">Say hello</button>
    <p id="greeting" role="status"></p>
    </html>
""".trimIndent()

Connect your platform's browser and pass its Page to the shared workflow. Call suspending APIs from a coroutine, on the main thread for mobile WebViews.

Browser(
  AndroidPlatformContext(
    activity, webView,
  ),
).use { browser ->
  demo(browser.page())
}

Browser(
  IOSPlatformContext(
    webView = webView,
  ),
).use { browser ->
  demo(browser.page())
}

launchBrowser(
  executablePath,
).use { launched ->
  demo(
    launched.browser.newPage(),
  )
}

Features follow each backend's abilities. Check page.capabilities() before an optional feature; the capability matrix lists what is partial or missing.

Expose runCommand() as a tool in your LLM SDK. The model reads a snapshot and acts on element refs, using the agent-browser command subset.

page.runCommand("snapshot -i")
// @e1 [input type="email"]
// @e2 [button] "Sign in"

page.runCommand("fill @e1 ada@example.com")
page.runCommand("click @e2")

The example app puts a playground, recipes and an inspector on Android, iPhone and iPad.

• Find and assert. Roles, labels and text, with waiting built in. Locators
• Watch the network. Intercept requests and read the history. Networking
• Capture pages. Screenshots and image comparison. Screenshots
• Keep a session. Save and restore cookies and storage. Session state
• Record a run. Video, traces and CPU profiles on Chromium. Recording

KMPwright adapts browser behavior from agent-browser into Kotlin APIs and is independent of Playwright. The upstream comparison records the pinned revision and coverage; no upstream source is included here.

Part of Kotlin Reforge.

_{Contributing? See development and tests and the release guide. MIT · Third-party notices}