Compose UI Control
OrganizationJetBrains
Control a running Compose Desktop application via HTTP. Use when you need to interact with UI elements, click buttons, enter text, wait for elements to appear, or capture screenshots in a Compose Desktop app that has compose-ui-test-server enabled.
Overview
- Repository
- JetBrains/skills
- Publisher
- JetBrains
- Stars
- 394
- Forks
- 35
- Bundled files
- 25
- License
- Not stated
- Updated
- Oct 21, 2026
Installation
tmp=$(mktemp -d)
git clone --depth 1 https://github.com/JetBrains/skills.git "$tmp"
mkdir -p ~/.claude/skills
cp -R "$tmp/compose-ui-test-server" ~/.claude/skills/compose-ui-control
rm -rf "$tmp"tmp=$(mktemp -d)
git clone --depth 1 https://github.com/JetBrains/skills.git "$tmp"
mkdir -p ~/.codex/skills
cp -R "$tmp/compose-ui-test-server" ~/.codex/skills/compose-ui-control
rm -rf "$tmp"Install the Compose UI Control agent skill from https://github.com/JetBrains/skills/tree/main/compose-ui-test-server into this project's skills folder.Instructions
compose-ui-test-server
A library that enables AI coding agents to control Compose Desktop applications at runtime via HTTP.
- Repository: https://github.com/forketyfork/compose-ui-test-server
- Maven Central:
io.github.forketyfork:compose-ui-test-server - Current version: 0.2.0
Checking If Already Installed
Before setting up, check if the project already has the library:
grep -r "compose-ui-test-server\|composeuittest" --include="*.gradle*" --include="*.kt" .Look for:
- Dependency on
io.github.forketyfork:compose-ui-test-server - Imports from
io.github.forketyfork.composeuittest
If found, skip to Starting the Application.
Installing in a Compose Desktop Project
Step 1: Add the dependency
Find the app's build.gradle.kts and locate the desktop source set dependencies. Add compose-ui-test-server and compose.uiTest:
kotlin {
sourceSets {
val desktopMain by getting {
dependencies {
// Existing dependencies...
// Add these two:
implementation("io.github.forketyfork:compose-ui-test-server:0.2.0")
@OptIn(org.jetbrains.compose.ExperimentalComposeLibrary::class)
implementation(compose.uiTest)
}
}
}
}For projects using version catalogs, add to gradle/libs.versions.toml:
[libraries]
compose-ui-test-server = { module = "io.github.forketyfork:compose-ui-test-server", version = "0.2.0" }Then reference in build.gradle.kts:
implementation(libs.compose.ui.test.server)Step 2: Update the main function
Find the application's main() function (usually in Main.kt or similar). Replace the standard Compose Desktop launcher with runApplication:
Before (typical Compose Desktop main):
import androidx.compose.ui.window.Window
import androidx.compose.ui.window.application
fun main() = application {
Window(onCloseRequest = ::exitApplication, title = "My App") {
App()
}
}After (with agent control support):
import io.github.forketyfork.composeuittest.WindowConfig
import io.github.forketyfork.composeuittest.runApplication
fun main() =
runApplication(
windowConfig = WindowConfig(
title = "My App",
minimumWidth = 1024,
minimumHeight = 768,
),
) {
App()
}The app now runs normally by default, but supports agent control when launched with COMPOSE_UI_TEST_SERVER_ENABLED=true.
Step 3: Add test tags to UI elements
For agents to interact with specific UI elements, add test tags:
import androidx.compose.ui.Modifier
import androidx.compose.ui.platform.testTag
Button(
onClick = { /* ... */ },
modifier = Modifier.testTag("login_button")
) {
Text("Login")
}
TextField(
value = username,
onValueChange = { username = it },
modifier = Modifier.testTag("username_field")
)Starting the Application
# Normal mode (no server)
./gradlew run
# Agent-controlled mode (server enabled)
COMPOSE_UI_TEST_SERVER_ENABLED=true ./gradlew run
# With custom port (default is 54345)
COMPOSE_UI_TEST_SERVER_ENABLED=true COMPOSE_UI_TEST_SERVER_PORT=8080 ./gradlew runVerifying the Server
Always check health first:
curl http://localhost:54345/health
# Expected: "OK"Available Endpoints
| Endpoint | Description |
|---|---|
GET /health |
Health check |
GET /onNodeWithTag/{tag}/performClick |
Click element by test tag |
GET /onNodeWithTag/{tag}/performTextInput?text=... |
Enter text (URL-encode the text!) |
GET /onNodeWithText/{text}/performClick |
Click element by display text |
GET /waitUntilExactlyOneExists/tag/{tag}?timeout=5000 |
Wait for element by tag |
GET /waitUntilExactlyOneExists/text/{text}?exact=true&timeout=5000 |
Wait for element by text |
GET /waitForIdle |
Wait for UI to stabilize |
GET /captureScreenshot?path=screenshot.png |
Capture screenshot |
Workflow Pattern
Follow this sequence for reliable interactions:
# 1. Verify server is running
curl http://localhost:54345/health
# 2. Wait for UI to be ready
curl http://localhost:54345/waitForIdle
# 3. Perform action
curl "http://localhost:54345/onNodeWithTag/username/performTextInput?text=myuser"
# 4. Wait for UI to settle
curl http://localhost:54345/waitForIdle
# 5. Perform next action
curl http://localhost:54345/onNodeWithTag/login_button/performClick
# 6. Wait for result
curl "http://localhost:54345/waitUntilExactlyOneExists/tag/dashboard?timeout=10000"
# 7. Capture screenshot to verify
curl "http://localhost:54345/captureScreenshot?path=result.png"Finding Test Tags
Search the codebase for existing test tags:
grep -r "testTag\|Modifier.testTag" --include="*.kt" .Also check:
CLAUDE.mdfor documented test tags- Test files in
src/*Test/directories
Important Notes
- Always URL-encode special characters in text: space→
%20,@→%40,&→%26 - Use
waitForIdlebetween operations for stability - Check HTTP status codes: 200=success, 400=bad request, 500=error
- Use appropriate timeouts for waits (default 5000ms may be too short)
- Screenshots require absolute paths
Error Handling
If an endpoint returns an error:
- Check the element exists (search for its test tag in code)
- Ensure UI has finished loading (
waitForIdle) - Verify the server is still running (
/health) - Try with a longer timeout for wait operations
Bundled files
.github/dependabot.yml342 B.github/workflows/build.yml646 B.github/workflows/junie.yml472 B.github/workflows/release.yml2 KB.gitignore343 BCLAUDE.md3 KBLICENSE2 KBREADME.md9 KBbuild.gradle.kts3 KBdocs/PUBLICATION.md7 KBgradle/libs.versions.toml909 Bgradle/wrapper/gradle-wrapper.properties252 Bgradlew9 KBgradlew.bat3 KBsettings.gradle.kts145 Bsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/ApplicationLauncher.kt3 KBsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/ComposeUiTestServer.kt5 KBsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/ComposeUiTestServerConfig.kt2 KBsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/Extensions.kt4 KBsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/TestEndpoint.kt2 KBsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/endpoint/ClickEndpoints.kt3 KBsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/endpoint/HealthEndpoint.kt611 Bsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/endpoint/ScreenshotEndpoint.kt2 KBsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/endpoint/TextInputEndpoint.kt2 KBsrc/jvmMain/kotlin/io/github/forketyfork/composeuittest/endpoint/WaitEndpoints.kt4 KB
FAQ
What does Compose UI Control do?
Control a running Compose Desktop application via HTTP. Use when you need to interact with UI elements, click buttons, enter text, wait for elements to appear, or capture screenshots in a Compose Desktop app that has compose-ui-test-server enabled.
Who made Compose UI Control?
JetBrains (an organization) publishes it in the JetBrains/skills repository on GitHub.
Is Compose UI Control free?
Compose UI Control is public on GitHub, but its repository has no license, so check with JetBrains before reusing it in your own work.
What comes with Compose UI Control?
Its instructions and 25 files the agent reads only when a task needs them, such as references, examples or scripts.
Which agents can use Compose UI Control?
Any coding agent that reads SKILL.md files, including Claude Code, Codex, Cursor, GitHub Copilot and Gemini CLI. The agent loads the skill when a task matches its description.
What are agent skills?
An agent skill is a folder with a SKILL.md file: instructions that teach a coding agent how to do one kind of task, and optionally files it can read while it works. The agent keeps only the skill's name and description in mind and reads the rest when a task calls for it.
How are agent skills different from MCP servers?
An MCP server gives an agent new tools to call, code that runs somewhere and returns results. A skill gives it knowledge: how to approach a task with the tools it already has. Many agents use both.
How can my agent test my Android app?
Robomori lets coding agents create, start and control Android emulators and devices from one command, so they can install your app, take screenshots and check dark mode, font sizes and languages.