Skip to content
RZ.

System 04

Developer Systems

CI/CD · MCP · QA Automation · AI Tooling

CI/CD, AI-assisted tooling, and QA automation built to keep a large team shipping fast without sacrificing release safety.

Components

AI-Assisted Engineering
Domain-specific CLAUDE.md AI context files introduced to accelerate team velocity.
CI/CD (Codemagic)
Tag-based build triggers and fixed Android store builds (NDK, Gradle cache, fastlane).
MCP Tooling
TypeScript MCP server exposing Codemagic as AI-tool APIs — trigger builds, check status, post Slack release notes.
QA Automation
Maestro automation for critical payment and banking flows.

Problem

As the team and app grew, build management, release notes, and QA coverage for payment-critical flows were manual and didn't scale with team size.

Constraints

  • Android store builds were breaking on NDK, Gradle cache, and fastlane issues, blocking releases.
  • Build status and release notes lived in Codemagic's UI, disconnected from where engineers and AI tools actually work.
  • Payment and banking flows are the highest-risk surface for regressions, so QA coverage there can't be an afterthought.

Architecture

Designed tag-based build triggers on Codemagic and fixed recurring Android store build failures across NDK, Gradle cache, and fastlane configuration.

Built a TypeScript MCP server exposing Codemagic as AI-tool APIs — trigger builds, check status, and post Slack release notes — with tag-based triggers and automated artifact paths for staging and production.

Drives test coverage strategy and Maestro automation for critical payment and banking flows, alongside domain-specific CLAUDE.md AI context files that give AI coding tools accurate project context.

Decisions

Expose CI as an API, not just a UI

Wrapped Codemagic in a TypeScript MCP server so builds, status checks, and Slack release notes are callable as AI-tool APIs instead of manual UI clicks, letting AI coding tools trigger and monitor builds directly.

Fix the build before automating around it

Android store builds were failing on NDK, Gradle cache, and fastlane issues. Fixed those root causes before layering tag-based triggers and MCP tooling on top, so automation wasn't automating a broken pipeline.

QA coverage where the risk is highest

Focused Maestro automation on critical payment and banking flows rather than broad UI coverage, since that is where a regression costs the most.

Impact

  • Built and shipped a TypeScript MCP server exposing Codemagic build management as AI-tool APIs.
  • Fixed Android store build failures across NDK, Gradle cache, and fastlane.
  • Driving Maestro automation for critical payment and banking flows.
  • Introduced domain-specific CLAUDE.md AI context files to accelerate team velocity.