Skip to main content

AI Integration

For AI assistants reading this page

The complete machine-readable reference is at https://docs.boozang.com/agent/AGENT.md — start there for concepts, workflows, and the full MCP API with parameters.

Boozang provides multiple ways for AI assistants to help with test automation.

Setup with bz-agent​

The bz-agent CLI tool is the recommended way to connect AI assistants to your Boozang project:

npx bz-agent init

This creates a .bz/ directory with:

  • AGENT.md — Entry point for AI assistants (give them this file)
  • config.json — Server URL, API endpoint, version
  • .env — Auth token (gitignored)
  • mcp-api.md — Full API reference with parameters for all 40+ tools
  • concepts.md, workflows.md, best-practices.md — AI-readable guides

Cache your project structure so AI agents don't need to re-fetch it:

npx bz-agent snapshot   # Creates .bz/project-map.md and modules-cache.json

Parse Boozang runner logs for test results and failures:

npx bz-agent parse worker1.log --pretty
npx bz-agent parse ./logs/ --build=1866

Conversation-Based Assistance​

The simplest way to use AI with Boozang is through conversation. AI assistants can:

Help Plan Tests​

Describe your application and user journeys. The AI will suggest:

  • How to structure modules
  • Which test types to use (functional, API, visual)
  • Data-driven testing strategies
  • Edge cases to consider

Guide Test Creation​

While working in Boozang, ask the AI for help with:

  • Element selection strategies
  • Handling dynamic content
  • Validation approaches
  • Control flow (loops, conditionals)

Debug Failures​

When tests fail, share the error details. The AI can:

  • Analyze failure patterns
  • Suggest root causes
  • Recommend fixes
  • Identify flaky test indicators

IDE Integration​

Modern IDEs with AI capabilities (Cursor, VS Code with Copilot) can assist when working with Boozang's coded tests or CI configurations.

Coded Tests​

For teams using Boozang's coded test feature, AI can help write JavaScript test code that integrates with Boozang's API.

CI Configuration​

AI can help configure:

  • Jenkins pipelines
  • GitHub Actions workflows
  • GitLab CI configurations
  • Docker-based test execution

MCP API Integration​

The Model Context Protocol (MCP) API enables programmatic AI integration:

AI Assistant ←→ MCP API ←→ Boozang Platform

Authentication uses project-scoped tokens (prefixed bzmcp_) with two scopes:

  • read — Browse projects, modules, tests, environments, IDE state
  • write — All read operations plus create, edit, delete, and IDE control

The API provides 40+ tools across six categories:

  • Module/Test/Action CRUD — Full lifecycle management
  • Environment management — Create, edit, provision environments
  • IDE control — Navigate, run tests, get screenshots, set breakpoints
  • Auth configuration — Manage authentication settings

See the MCP API Reference for an overview, or the full specification with parameters for complete details.

Best Practices for AI-Assisted Testing​

Be Specific​

Instead of: "Create tests for my app"

Try: "Create a login test module with tests for: valid login, invalid password, forgot password flow"

Provide Context​

Share relevant information:

  • Application URL and technology stack
  • User roles and permissions
  • Business rules and validation requirements
  • Known edge cases

Iterate​

AI-generated tests are a starting point. Review, run, and refine:

  1. Generate initial test structure
  2. Run tests to verify they work
  3. Adjust selectors or assertions as needed
  4. Add data variations

Maintain Human Oversight​

AI accelerates test creation but doesn't replace judgment:

  • Review generated tests for correctness
  • Validate business logic coverage
  • Ensure security-sensitive flows are properly tested
  • Monitor test reliability over time

Supported AI Assistants​

Boozang works with any AI assistant that can read the .bz/AGENT.md guide and make HTTP requests to the MCP API. The bz-agent init command optionally creates a CLAUDE.md pointer file for Claude Code.

Popular options include:

  • Claude Code - Anthropic's CLI agent (reads CLAUDE.md automatically)
  • Cursor - AI-powered code editor
  • ChatGPT - OpenAI's conversational AI
  • GitHub Copilot - AI pair programmer