- Category
- AI Agents
- Rank
- No. 1452Tools index
Previous survey · No. 1459 ·
- Pricing
- Open Source
- Type
- TOOL
- Builder
- browser-use
- GitHub
- 829 stars
- Date
About
MCP server that runs automated QA on your web app using Browser-Use agents. Cursor and Claude can crawl your site and report bugs.
What it does
Vibetest coordinates several Chromium sessions, gives each an exploratory QA task, and combines their observations into a structured report. It targets functional failures, broken navigation, accessibility problems, and performance trouble. You choose the site, agent count, and whether browsers remain visible. A second model pass deduplicates findings and sorts confirmed issues by severity.
Why it's ranked here
The design is useful for fast exploratory checks because parallel agents can exercise different parts of live or local sites, then consolidate their findings. The implementation also closes browser sessions and records individual failures. Still, this is a small, stateful prototype whose reports depend on model judgment rather than repeatable assertions.
What's good
Visible and headless runs support both debugging and unattended workflows. Concurrency is bounded at ten sessions, reducing runaway browser creation. Each agent records its assigned task, result, status, and timing, so partial failures remain visible. The consolidation prompt asks for exact elements, actions, outcomes, and context while rejecting subjective preferences and intentional omissions.
Tradeoffs
Results live only in process memory, so restarting the server loses every test identifier and report. Exploration uses a high model temperature, which makes repeated runs less deterministic. Security protections are disabled in the browser profile, and Chromium receives a no-sandbox flag. Installation requires Python, browser binaries, an MCP-capable client, and a Google API key. No automated tests appear in the supplied repository tree.
How to use it well
Use Vibetest during rapid web development, especially before manually checking a live preview or localhost build. Start with visible browsers to inspect agent behavior, then use headless mode for repeated exploratory passes. Treat reported defects as leads that need confirmation. Keep deterministic unit, integration, accessibility, and regression tests for release gates, because this tool does not provide stable assertions or persistent history.
Technical notes+
pyproject.toml defines a Python 3.11+ package and maps the vibetest-mcp console command to vibetest.mcp_server:run. vibetest/mcp_server.py exposes FastMCP tools named start and results, disables logging globally to protect JSON-RPC output, and disables anonymized telemetry. vibetest/agents.py runs Browser-Use sessions under an asyncio.Semaphore capped at ten, stores run data in the module-level _test_results dictionary, and uses Google chat models for exploration and severity summarization. Browser profiles set disable_security=True and pass --no-sandbox. vibetest.egg-info/SOURCES.txt lists package metadata and three Python source files, with no test directory or test files listed.
Observed
- License
- MIT
- Primary language
- Python
- Python requirement
- Python 3.11 or newer
- Install surface
- Python package with a documented editable uv installation and separate Playwright Chromium installation
- Interface
- MCP server exposed through the vibetest-mcp console command
- Supported targets
- Live websites and localhost development sites
- Browser runtime
- Playwright Chromium through Browser-Use
- Repository structure
- The supplied source manifest lists no test directory or test files
Read from README.md, pyproject.toml, vibetest/agents.py, vibetest/__init__.py, vibetest/mcp_server.py, vibetest.egg-info/PKG-INFO, vibetest.egg-info/SOURCES.txt, vibetest.egg-info/requires.txt, vibetest.egg-info/top_level.txt, vibetest.egg-info/entry_points.txt, vibetest.egg-info/dependency_links.txt.
What it can do
Crawl web application pages automatically
Web application URL → Site structure and page content data
Execute automated quality assurance testing
Web application → QA test results and findings
Detect and identify bugs in web applications
Web application pages and functionality → Bug reports with specific issues found
Generate automated test reports
QA testing results → Formatted bug and issue reports
Integrate with Cursor and Claude development tools
MCP server connection → Connected testing environment
Run browser-based testing agents
Web application targets → Automated browser test execution results
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.