MCP serverfor Claude Code, Cursor & Windsurf

Smoke tests that
run themselves.

VibeCheck MCP connects your AI coding agent to a real Playwright browser. Describe a flow in plain English — the agent clicks through your app by itself, validates the result, and every run is recorded: video, console, network, every step. Failures arrive as ready-made, shareable bug reports.

Free API key · Works on localhost & VPN staging · No Playwright code to write

app.vibecheck-qa.com/tracks/1783009013 — smoke: checkout flow
0:08 / 0:12
0:01Navigatestaging.myapp.com/cart
0:03Clickbutton “Add to cart”
0:05Typecoupon → “SAVE20”
0:08Validatetotal updated to $33.60 ✓
0:11Finishrecording uploaded → shareable link
12s · video + console + network + 5 actions — share with anyoneSee the prompt
Playwright-driven, Real Chromium — watch it work, Smoke tests in plain English, Video + console + network, Works on localhost, Claude Code, Cursor, Windsurf, Shareable track links, Failures = ready-made bug reports

How it works

Prompt → browser → recording.

Two minutes of setup. Then your agent tests anything you can reach — localhost included.

1

Add the MCP server

One command in Claude Code (or a JSON snippet for Cursor, Windsurf, VS Code). Grab a free vck_ API key from Settings → API Keys.

$ claude mcp add vibecheck …
VIBECHECK_API_KEY=vck_…
2

Describe the test

Plain English, no Playwright code: “Go to /checkout, add a product, verify the total updates — record it and send me the link.”

“Go to /checkout, apply SAVE20, verify the total drops 20%. Record it.”

3

Get the recording

The agent drives a visible Chromium on your machine, validates the flow, and replies with a shareable track — pass or fail.

…/tracks/1783009013
passed ✓12s · 5 actions

Agent-driven QA

You describe the flow. The agent does the rest.

No test scripts, no selectors, no CI setup. Your coding agent already understands your app — VibeCheck MCP gives it a real Playwright browser and records everything it does. Run smoke tests after every change, before every deploy, or whenever a PM asks “is checkout still working?”

  • Plain-English tasks — the agent translates them into real clicks
  • Visible Chromium on your machine: localhost and VPN staging work
  • Validates results and reports pass or fail, with proof
  • Every run is a track: video, console, network, action timeline
  • A failed run is already a bug report — share the link, or let AI Fix open the PR
Set up Agent Recording
claude code

Go to staging.myapp.com/cart, apply coupon SAVE20 and check the total drops 20%. Record it and send me the link.

Pricing

Start free. Scale when you ship.

The MCP server is open source. Agent runs count as recordings — 25 a month free, no credit card.

Free

For trying agent-driven QA on your own project.

$0 / month
  • 25 recordings / month (agent or human)
  • MCP server + Agent Recording
  • 3 AI Fixes / month
  • 7-day retention
  • Repro Links
Get started free

Pro

Most popular

For teams running smoke tests every day.

$12 / user / month
  • Unlimited recordings & agent runs
  • 50 AI Fixes / month
  • 90-day retention
  • Production error monitoring (Watch SDK)
  • Quick Fix for non-devs
  • Slack & Linear handoff
Start Pro

Enterprise

BYO storage

For teams with compliance and a real legal review.

Custom
  • Bring-your-own storage (S3/GCS/Azure/R2)
  • Unlimited AI Fixes & retention
  • SAML SSO + SCIM
  • Custom DPA, SOC 2 reports
  • Dedicated Slack channel
Talk to founder

Frequently asked questions

Agent Recording lets AI coding agents like Claude Code, Cursor and Windsurf drive a real browser on your machine via the VibeCheck MCP server. You describe a flow in plain English; the agent performs it with Playwright while VibeCheck records the whole session — video, console logs, network requests and a labeled action timeline — and returns a shareable track link.

Your agent writes the code.
Let it test the app too.

One command to install. One prompt to run. One link as proof.