MockMouse

Turn natural language into intelligent web tests - no code required

Comment

MockMouse: Cursor for QA

Inspiration

It was frustrating spending half of HackDavis' time manually clicking through our app instead of building features. At 3 AM, while other teams were adding cool features, we were stuck testing feature flows for the 50th time. That's why we built MockMouse - an AI agent that tests your web apps while you focus on what actually matters. The inspiration for MockMouse came from a simple yet frustrating observation: testing web applications shouldn’t require a computer science degree.

We saw:

  • Product managers and designers struggling with test scripts
  • QA engineers overwhelmed by repetitive tasks

“Login with valid credentials and check if the dashboard loads.”
Computers don't understand this. We make them.

What It Does

MockMouse is an AI-powered application testing platform that:

  • Accepts natural language test instructions
  • Records and breaks down tests step by step
  • Uses AI to explore and understand web apps
  • Streams test runs live with visual feedback

How We Built It

Architecture Overview
MockMouse is built using a microservices-based structure with:

  • Frontend (Next.js)
  • Backend API (Express)
  • AI Testing Infrastructure (Python, Browser-use)

Technology Stack

Frontend

  • Next.js 14 + TypeScript
  • Socket.IO Client for real-time streams
  • Tailwind CSS for clean, modern UI
  • Custom components for test and stream visuals

AI Layer (Python)

  • Browser-Use library for automation
  • OpenAI / Anthropic APIs for natural language understanding
  • Socket.IO for real-time communication

Backend API

  • Python + SocketIO for API + WebSocket support
  • Supabase + Prisma for persistence
  • Authentication via Clerk

Key Features

Natural Language Test Creation
Write tests like: "Login with valid credentials and check if dashboard loads."

Intelligent App Discovery
AI automatically maps apps by:

  • Crawling pages
  • Detecting forms and UI elements
  • Building a knowledge graph of user flows
  • Recognizing login/search/checkout patterns

What We Learned

Technical Discoveries

  • AI Agent Design for app flow understanding
  • Real-time streaming with low latency
  • Handling async operations + lazy loads
  • Teaching AI workflows and user intent

Product Insights

  • User experience is everything
  • Visual feedback builds trust
  • Simplicity > Complexity, but with flexibility

Team Takeaways

  • Cross-functional dev is critical
  • MVPs validate assumptions early
  • Design with user testing from day one

Technical Challenges

Agentic Flow with browser_use
Challenge: Designing a reusable AI loop that can navigate and test web apps using visual understanding
Solution:

  • Built multi-step agentic workflows for planning and execution
  • Leveraged browser-use for screen-based visual inference
  • Implemented memory and context retention across test steps
  • Adapted flow to handle async behavior, modals, and visual state changes

Dynamic UIs Break AI
Challenge: Web apps change constantly
Solution:

  • Retry logic with exponential backoff
  • Context-aware detection models
  • Fallbacks + adaptive wait logic

Browser Automation at Scale
Challenge: Avoid memory leaks & crashes
Solution:

  • Browser lifecycle control
  • Session isolation
  • Auto-recovery from crashes

Product Challenges

Simplicity vs Power
Challenge: Powerful AI without a scary UI
Solution:

  • Natural language test input
  • Visual feedback for all steps
  • Guided onboarding and examples

Earning Trust in AI
Challenge: Users need to understand AI actions
Solution:

  • Real-time reasoning logs
  • Visual AI confidence levels
  • Post-test replay mode

Team Challenges

Tight Timelines

  • Focused on MVP
  • Used parallel dev streams
  • Standardized code & docs

Complex Integration

  • Clear API endpoints
  • Shared TypeScript interfaces

What We’re Proud Of

Technical Wins

  • Zero-downtime Python Infra
  • Under 1s latency test streaming
  • 99% uptime through fault recovery
  • Scalable infra for multiple sessions

Product Highlights

  • Easy-to-use UX for non-tech users
  • Real-time, interactive test visuals
  • Clear docs with walkthroughs

Innovations

  • Natural Language Testing
  • Context-Aware AI Agents
  • Replay Mode + Visual Logs

Built with love by Team MockMouse for Berkeley Hacks 2025

Built With

Share this project:

Updates