Labsco
MCP SERVER

Limetest

by m2rads

Describe an end-to-end test in plain language and let the model drive a real Chrome through it — Playwright snapshots first, vision only when those fall short.

Automated Testing & QA
Summary
Snapshots over screenshots, which is why it is quick.

Driving a browser by accessibility snapshot instead of pixels is both cheaper per step and less brittle when a layout shifts, and the vision fallback covers what the snapshot genuinely cannot see. The separate profile deserves a moment's thought before you point it at a staging login: session state persists across runs, which is convenient until it is not.

What it is

Limetest, a lightweight end-to-end testing framework with an MCP server built on Playwright. Test cases are written in natural language and executed by a model against a real browser; execution leans on Playwright's accessibility snapshot rather than pixel analysis, which is what keeps it fast and stable, and falls back to vision mode for scenarios the snapshot cannot express.

What you get
  • End-to-end test runs planned and executed by the model against a real Chrome instance
  • Snapshot-driven execution, with a vision fallback for the cases snapshot mode cannot handle
  • A headless mode for running the same tests inside CI
  • A persistent Chrome profile, so a login survives between sessions until you delete it
Requirements

An OpenAI API key — the README states it is needed to run this server in end-to-end mode, and it is passed on the launch command as `--api-key`. Install `@limetest/mcp` from npm and run `npx playwright install` to fetch the browser. The server launches Chrome under its own profile, stored under `ms-limetest/mcp-chrome-profile` in the platform cache directory; anything you log into stays there until that folder is removed. Apache 2.0, and built on Microsoft's Playwright MCP.

Setup effort

One command plus a key — npm install @limetest/mcp, then supply credentials