Introduction
Hover is an open-source Vibe Testing suite built around an MCP server (@hover-dev/mcp). Add it to the coding agent you already run (Claude Code, Cursor, …) and the agent explores your app, maps its business flows, and crystallizes each one into a standard Playwright spec under __vibe_tests__/. The saved tests are yours — they run in your CI with zero AI in the loop.
The differentiator is record == replay: the agent acts through Hover's grounded browser tools, so the selector that drove a click is the exact one saved, and crystallization is deterministic (no LLM writing code). No confabulated selectors, no runtime dependency on Hover, no lock-in.
Nothing changes in your app, your bundler config, or your dependencies — Hover drives your existing dev server over CDP regardless of framework (Vite, Next.js, Nuxt, Astro, webpack).
The suite — four surfaces, one artifact
| Surface | Role | What it is |
|---|---|---|
MCP — @hover-dev/mcp | author | The engine. Add it to your own agent; /mcp__hover__test_app explores + crystallizes specs. BYO-CLI — your model, your subscription. |
VS Code — hover-dev | review | An optional cockpit: a Business Map graph of your flows + coverage, and a Dashboard (pass / fail / flaky + CI results). |
| CI | run | The crystallized specs run on every PR as plain Playwright — no agent, no tokens. Hover generates the workflow. |
| Cloud (beta) | watch | Scheduled monitoring, a flakiness dashboard, regression alerts, on-failure self-heal orchestration — over the specs you already own. Runs in your CI, not ours. |
The through-line is the artifact: owned, portable Playwright in your repo and your CI. The AI authors it once; nothing AI runs after.
What's inside this section
- Quick start — Add the MCP and run your first
/mcp__hover__test_app. 60 seconds. - Install — Add
@hover-dev/mcpto your agent; the optional VS Code cockpit. - Your first session — A guided walkthrough on a real flow.
- Pick an agent — Which coding agents work and how Hover plugs in.
At a glance
- No API key required. Hover bundles no model and resells no tokens. It rides the coding agent and subscription you already pay for (BYO-CLI); your agent handles its own auth.
- Nothing to wire into your app. Hover drives your existing dev server over CDP — no plugin, no config edit, no new dependency. Works with any framework.
- record == replay. The agent acts through grounded tools (
role+name → testId → text), so the selector that drove an action is the exact one saved; crystallization is deterministic, no LLM writing code. - One artefact, plain Playwright. Each verified flow is a
@playwright/testspec that checks into git and runs in CI with zero AI, no tokens, no key.
Do I need the VS Code extension?
No. The MCP is the whole authoring loop. The VS Code extension is an optional review cockpit — a Business Map graph + a Dashboard — that drives no agent.
Building Hover itself? See Development for the monorepo workflow.