AI & MCP Automation Overview
TestFly has two distinct AI paths:
- Java runtime:
act(),byIntent(), AI assertions, action caching and patch generation run inside TestFly tests. Configure the Java framework's AI provider separately from your IDE assistant. - External MCP tooling: an assistant uses Playwright MCP for browser inspection and a separate Node.js TestFly bridge for project/code generation and cache/patch operations.
Assistant → Playwright MCP → browser inspection
→ TestFly Node bridge → Java source / cache / patch tools
Java tests → configured AI provider → runtime actions / assertions
Source checked on 2026-10-04: package.json declares @testfly/mcp 1.1.0, Node.js 18+, and bin/testfly-mcp.js. The public npm registry returned 404 for that package, so this guide uses the source checkout. The package name/version in source is not proof of an npm release.
Workflow
Inspect the real application DOM before choosing locators. Prefer semantic locators supported by io.testfly.locator. Review generated code and compile it against your TestFly dependency; code generation does not prove that a test works. Run the test against a controlled environment before using it in CI.
Components
| Component | Where to start |
|---|---|
| Java runtime AI features | Agentic testing |
| Six-tool Node MCP bridge | MCP Bridge & CLI |
| Live browser inspection | Microsoft Playwright MCP |
| IDE source projects | IDE plugins |
| Generation prompts | Prompt recipes |
The earlier Python/Selenium server and 88-tool recorder model are historical. The ADR, Web Studio and recorder pages preserve that design without advertising those commands as current Node features.