Mert Erden — Senior iOS Engineer | Developer Tools & AI Pipelines

Projects

What I'm building

M

MCP Servers & Tools

Model Context Protocol servers that give AI editors eyes and hands on mobile simulators. Install with npx, connect to your editor, and let AI see your app's screen, tap buttons, run accessibility audits, and validate store compliance.

v2.2.0Stable

Dev Toolkit MCP

59 MCP tools -give your AI editor eyes and hands on iOS & Android simulators

MCP server bridging AI editors with mobile simulators.

•See, interact with, and audit iOS + Android apps from your editor
•42 device control tools: screenshot, tap, swipe, type, dark mode, locale, permissions
•Autonomous agent DSL for multi-step flows in a single MCP call
•Accessibility audit and Store compliance checks
•Supports Claude Code, Cursor, Windsurf, Copilot CLI, Cline, Claude Desktop

App Store Screenshots

AI skill that builds custom screenshot studios -generative, not templated. 21 devices, 7 frame types, 11 slide types

AI skill that builds custom screenshot generators from scratch.

•900+ line SKILL.md teaches AI assistants the full workflow
•Every project gets unique, tailored JSX - generative, not templated
•7 CSS-only device frames across 21 resolutions (iPhone, iPad, Android, Watch)
•11 slide types including marketing-only (Award Hero, Testimonial, Stats)
•Marketing copy engine with narrative arc (Scene, Result, Relief)
•4 switchable themes, multi-language with RTL, browser-side PNG export

Next.jshtml-to-imageSKILL.md (900+ lines)CSS Device Frames

P

AI Pipeline

End-to-end development automation: from a Jira ticket or GitHub issue to a reviewed, tested, and committed PR -driven by multiple AI models working in sequence with deterministic quality gates.

v5.2.1Stable

Multi-Agent Pipeline

8-phase AI pipeline · Phase 7 Report (Jira · Wiki · Confluence · report) · Phase 0 URL enrichment (Firebase Crashlytics + Fortify) · skill security scanner · log search · generic figma-to-component (iOS + Android) · 37 smoke suites · local-only CI (pre-push hook) · plan-driven source sync

End-to-end AI development pipeline (v5.2.1).

•8-phase flow: analysis → planning → TDD dev → 3-model review → test → commit/PR → Report
•Phase 7 Report consolidates all external delivery + internal capture into one logical phase: Step 1 humanizer'd Jira comment (analysis + test scenarios) → Step 2 component wiki + Figma screenshots (4-adapter: submodule / in-repo / github-wiki / separate-repo) → Step 3 optional Confluence page → Step 4 agent-log + Quality & Metrics + telemetry → Step 5 knowledge capture (architecture, patterns, gotchas, decisions) + memory updates
•Opus + GPT-5.4 + Sonnet review in parallel, Opus triage filters false-positives
•v5.1.0 skill security scanner: tiered pattern catalog (critical/high/medium/low) running as pre-deploy hook on install.js + CI --strict + /multi-agent:scan slash, catches shell-pipe exec, base64 exec, eval-of-network, unicode bidi override (trojan source), hardcoded credentials with FORBIDDEN-context whitelist
•v5.1.0 log search: /multi-agent:search full-text across ~/.claude/logs/multi-agent/, smart ranking (matches × recency), filters --project/--since/--phase, ripgrep primary with grep fallback
•v5: generic figma-to-component pipeline for iOS (SwiftUI) and Android (Jetpack Compose) with the same workflow — 5 platform-specific skills under figma-ios/figma-android + 27 shared skills under figma-common, idempotent placeholder transform, overlay protection for locally-enhanced skills
•issue → jira → wiki triad: GitHub issue auto-creates Jira (configurable), Phase 7 Step 2 posts humanizer'd wiki summary back to Jira
•corporate hosts (Jira/Confluence/Bitbucket) prompted once per user during Token Save Flow Step 3.5, stored in ~/.claude/multi-agent-preferences.json (never committed), tokens always in Keychain
•live progress-line contract: every phase emits → <verb> <object> at immediate flush, progressVerbosity (quiet/normal/verbose)
•plan-driven source sync — --plan writes JSON artifact listing every rename + target tree + overlay-skip before apply
•multi-repo mode with per-repo worktrees, platform identity routing, push-must-succeed (5 retries), fetch-fail 4-option flow
•bilingual setup (en/tr, Step 0 language picker)
•cross-CLI full parity across 20 commands + 32 figma skills — byte-identical behaviour between Claude Code and Copilot CLI
•36 smoke suites, ≥506 assertions, personal-data leak gate (13 patterns, case-insensitive) always green
•Opt-in telemetry, pre-commit secret scan, 6 sub-agents, cross-session memory

Node.jsClaude CodeCopilot CLI177 Unified SkillsGitHub PackagesJSON Schemas6 Sub-Agents

v1.5.0Stable

Remote Control

Pocket Dev Environment — control your Mac from your phone

Full development toolkit on your phone over an encrypted relay.

•AI chat (Copilot CLI + Claude CLI), shell, file browser, git ops
•Build/test/lint with auto project detection (Node, Swift, Xcode)
•System monitor: CPU, RAM, disk, battery, Xcode, ports
•Multi-agent pipeline v3.7.0 with 18 commands, multi-repo orchestration, identity routing
•Audit log, secret redaction, blocklist v2, safe mode
•AES-256-GCM encryption via private GitHub Gist
•Path-jailed file operations — only whitelisted project dirs
•No tunnels or open ports — works on company laptops