The endless loop of fixing brittle end-to-end tests after minor frontend tweaks has a new challenger. Pluswithai has introduced an autonomous AI agent designed to explore web applications and generate standard Playwright TypeScript code, specifically targeting the pain points of flaky CI pipelines. By leveraging the Model Context Protocol (MCP), this tool integrates directly into developer workflows, offering a bridge between agentic exploration and local code execution without trapping users in proprietary black-box runners.

The Three Core Bottlenecks of Modern E2E Testing

Modern E2E testing suffers from three main issues: brittle locators that break when CSS classes change, scaffolding friction where writing boilerplate for auth flows consumes engineering hours, and the vendor lock-in trap where tests are hidden inside proprietary cloud dashboards. If you cancel a subscription to a black-box tool, you often lose your entire test suite.

The Agentic Approach Over Traditional Recording

Unlike traditional recorders that capture raw coordinates or fragile CSS selectors, this agent mimics human QA behavior by analyzing the page's Accessibility Tree. It identifies interactive elements through functional roles and ARIA attributes, ensuring that locators like page.getByRole() and page.getByLabel() are used instead of arbitrary class names. This semantic inspection allows the agent to map user journeys dynamically, discovering signup flows, navigation states, and checkout steps autonomously.

Integration via CLI and MCP Servers

The architecture explicitly avoids vendor lock-in by emitting idiomatic Playwright code that can be committed directly to Git repositories. Developers can interact with the system through a CLI tool (@pluswithai/cli) or via an MCP server that connects to environments like Cursor and Claude Desktop. This setup allows engineers to prompt their IDEs to 'explore my staging app and generate a Playwright suite,' receiving executable code directly in their editor rather than viewing results in an isolated dashboard.

Current Engineering Challenges and Open Trials

While the core generation engine is functional, the team acknowledges ongoing hurdles in handling complex edge cases like multi-factor authentication (MFA), single-sign-on (SSO), and realistic mock database states for destructive operations. The project is currently in an open trial phase with no credit card requirement, inviting feedback on how the agent handles specific application flows. Future development focuses on auto-healing selectors when fundamental UI redesigns occur, aiming to maintain test stability even as the underlying DOM structure shifts.

Key Takeaways

  • The agent uses Accessibility Tree analysis to generate resilient Playwright locators, avoiding brittle CSS class dependencies.
  • Output is standard TypeScript code compatible with existing GitHub Actions and local runners, preventing vendor lock-in.
  • MCP integration allows direct interaction from IDEs like Cursor and Claude Desktop, keeping testing within the developer loop.
  • The tool is currently in an open trial, actively seeking feedback on MFA, SSO, and database state management challenges.

The Bottom Line

By decoupling test generation from proprietary runners and focusing on semantic DOM inspection, Pluswithai addresses the root causes of E2E flake rather than just patching the symptoms. This approach respects developer workflows by keeping code in the repo, making it a pragmatic evolution for teams tired of selector maintenance.