Autonomous browser testing

Agents test your app in real browsers.

Describe the flow — or wire the Glidepath MCP into Cursor or Claude Code. Agents drive Chromium, Firefox and WebKit, handle signup emails, OTPs and 2FA, and land a verdict with a replay, artifacts, and an exported Playwright test.

Get started free

Free tier: 2 local agents, all 32 MCP tools. No card required.

agent run · shop.example.com
$glidepath run --url shop.example.com --browsers chromium,firefox,webkit
plan3 flows discovered — signup, checkout, password-reset
chromiumsignup · verification code pulled from disposable inboxPASS
firefoxcheckout · cart total asserted $42.00PASS
webkitpassword-reset · reset link followed from emailPASS
verdict3/3 flows · replay + trace saved · Playwright test exported

How it works

From prompt to verdict, unattended.

01

Connect

Start a run from the dashboard or the API — or let your coding agent do it over MCP, with 32 browser tools and device-code auth.

02

Agents fly the flows

Real Chromium, Firefox and WebKit at desktop and mobile viewports. Disposable inboxes and SMS numbers for verification codes, live TOTP for 2FA, saved accounts for login.

03

Land the verdict

Pass or fail with a failure class, step-by-step events, video and trace artifacts, session replay — and a Playwright spec you can commit.

Features

The whole approach, not just the click path.

MCP for coding agents

32 tools over stdio. Cursor, Claude Code or Claude Desktop drive real browsers — navigate, click, assert, screenshot — on your quota, with BYOK.

Cross-browser matrix

Every run fans out across engines and viewports with one verdict per cell, so “works in Chrome” stops being the bar.

Auth helpers built in

Disposable email and SMS inboxes for verification codes, live TOTP for 2FA, and a vaulted account store agents can log in with.

Human tester sessions

Send a hosted copy of your app to anyone. Their clicks, findings and narrated voiceover come back as a replay with transcript.

Findings inbox

Agent and tester findings land in one queue. Triage, dismiss — or generate a repro test from a finding in one click.

API test sweeps

Point at an OpenAPI spec or a base URL; agents exercise the endpoints and report results per case.

Quickstart

Two minutes to the first run.

The MCP server runs on your machine and authenticates with a device code — no API keys pasted into terminals. Hosted execution needs nothing installed at all.

$ npm install -g browser-automation-mcp

$ browser-automation-mcp setup

→ opens app.glidepath.to/cli/connect · approve with code WGHX-9TKP

connected — your agent now has 32 browser tools

Pricing

Free to fly locally. Pay for the cloud.

Free

$0

Everything you need to test locally.

  • 2 local agents
  • All 32 MCP tools
  • Findings inbox
  • Device-code CLI auth
Start free
Most popular

Pro

$29/mo

Hosted execution for teams shipping weekly.

  • 10 concurrent agents
  • Hosted cloud execution
  • 500 agent-minutes included
  • *.glidepath.to tunnel domain
  • $249/yr — two months free
Upgrade to Pro

Pay as you go

$0.05/agent-min

No commitment, on top of any plan.

  • Available on every plan
  • Billed at period end
  • Same hosted infrastructure
Get started
Glidepath — autonomous browser agents that test your web app