Limetest

Limetest MCP Server enables automated, AI-powered end-to-end testing within MCP client workflows. It leverages Playwright under the hood to validate changes across your app through natural-language test definitions and efficient execution, making CI validation faster and more reliable.

How to use

You can run end-to-end tests against your application by connecting your MCP client to the Limetest MCP Server. Use your API key to authorize the AI-driven test execution, and run tests in headless mode to integrate smoothly into CI pipelines. The server is designed to work with the Limetest client and your existing Playwright setup, providing end-to-end validation of changes in an automated, repeatable way.

How to install

Prerequisites you need before installing:
- Node.js (recommended LTS version) with npm
- Basic familiarity with MCP client configuration
- Internet access to install packages and run MCP commands

Install Limetest MCP and the necessary Playwright dependencies using npm and npx:

npm install @limetest/limetest

npx playwright install

Configure the MCP server for your MCP client. Use the following example to connect and provide your OpenAI API key. This setup uses a stdio approach, running the MCP server via npx with the required arguments:

{
  "mcpServers": {
    "limetest": {
      "command": "npx",
      "args": [
        "@limetest/mcp",
        "--api-key=<your openai api key>"
      ]
    }
  }
}

This config ensures all logged-in state is stored in the generated profile and can be cleared between sessions if you want to reset offline state.

## Additional configuration and notes

User data directory locations (where Limetest stores its browser profile) are:
```- %USERPROFILE%\AppData\Local\ms-limetest\mcp-chrome-profile (Windows)
- ~/Library/Caches/ms-limetest/mcp-chrome-profile (macOS)
- ~/.cache/ms-limetest/mcp-chrome-profile (Linux)```
These profiles help preserve state between test runs. You may delete them to clear offline state between sessions.

## Security and API key usage

Protect your API key and restrict access to your MCP client and Limetest MCP Server. Treat the API key as a sensitive credential and rotate it periodically. When running in CI, consider injecting the API key via a secure secret store or environment variable rather than embedding it directly in configuration files.

## Troubleshooting tips

If tests fail to run end-to-end:
- Verify that the Limetest MCP Server is reachable and the API key is correct.
- Ensure the Playwright dependencies are installed and that the browser can launch in your CI environment.
- Check the browser profile directory permissions to avoid write or read errors.
- If snapshot mode fails on a complex scenario, Limetest will automatically fall back to vision mode to continue validation.

## Tools and capabilities

Limetest provides several core capabilities for AI-driven testing, including end-to-end test execution, a lightweight snapshot-based verification path, and vision-based fallbacks for more sophisticated scenarios.

## Available tools

### ai\_execution

Executes end-to-end test scenarios described in natural language, coordinating with the MCP client and Limetest MCP Server to validate UI and API behavior.

### snapshot\_mode

Uses Playwright snapshot-based verification for fast, reliable checks and to avoid pixel-level analysis where unnecessary.

### vision\_fallback

Activates vision mode when snapshot verification fails, enabling broader scene understanding for complex UI interactions.