Show HN: I built E2E Test Agent – describe tests in plain English,AI executes it (github.com)

E2E Test Agent is an open-source framework that turns plain-English test files into runnable end-to-end checks by using LLM-powered agents to interpret intent and drive browser automation. Instead of writing fragile selector-based Playwright scripts, you write human-readable .test files (e.g., “open playwright.dev, scroll down, click ‘Get started’, verify sidebar”), and a TestAgent orchestrates an LLM Agent to plan actions and MCP tools (Playwright MCP server by default) to execute them. It’s installed via npm/yarn, configured with a .env (MODEL_NAME defaults to gpt-4o, API_KEY, BASE_URL, TESTS_DIR), and exposes a simple API (TestAgent.runAllTests(), runSingleTest()) that returns structured results (success flag, steps_completed, observations). This approach matters because it shifts testing from brittle selectors to intent-aware automation that can adapt to UI changes, be readable by non-developers, and integrate with multiple LLM backends (OpenAI, Anthropic via OpenAI-compatible endpoints, OpenRouter, local LLMs like Ollama/LM Studio). Key technical points: architecture = .test files → TestAgent → LLM planning → MCP executors (Playwright, filesystem, etc.); configurable maxSteps and MCP servers; outputs detailed JSON summaries. Practical implications include lower maintenance and potential self-healing tests, but also trade-offs around LLM reliability, cost, security, and reproducibility—areas the project invites contributors to improve (parallel runs, retries, screenshots, more MCPs).