Maneesh Projects & writing

Project / developer tools

DevTools MCP Server

DevTools MCP Server gives AI coding agents structured access to the development tools that normally stay on the developer machine: builds, simulators, emulators, tests, screenshots, logs, audits, and safe local automation.

View GitHub Back to projects

Purpose

Reduce manual friction in development by turning repeatable work into explicit tools an agent can call, verify, and report back from.

Direction

It started with iOS because that is where local automation is hardest. The direction is a multi-platform developer tool layer for Apple, Android, web, backend, CI, and release workflows.

Available now

Current tool surface.

iOS and Xcode

Build, inspect, run, and debug Apple-platform projects from an AI agent workflow.

  • ios_build runs xcodebuild and returns structured errors with file, line, column, and message.
  • ios_list_simulators lists simulator and device targets with UDID, state, and OS version.
  • ios_simulator_screenshot captures simulator screenshots for visual debugging.
  • ios_simulator_logs returns a bounded simulator log slice with optional filtering.
  • xcode_run_app, xcode_stop_app trigger Xcode run and stop through AppleScript.

Android

Gradle, emulator, device, test, lint, and Kotlin hygiene tools for Android projects.

  • android_build runs Gradle builds with structured error and warning parsing.
  • android_list_emulators, android_list_devices find available Android targets.
  • android_start_emulator, android_install_apk, android_run_app, android_stop_app handle emulator and app lifecycle work.
  • android_test runs unit or instrumented tests and parses JUnit results.
  • android_lint, audit_kotlin_hygiene catch Android Lint and Kotlin hygiene issues.

Workflow automation

Shared tools that make local development safer and repeatable across projects.

  • audit_changed_files checks changed files before commit and is designed to run beside builds.
  • list_projects exposes configured projects and whitelisted local CI commands.
  • run_ci runs only approved commands from project configuration.
  • gateway_build, gateway_start, gateway_stop manage the LLM Gateway during local development.
  • SSE mode lets container-based agents call host-machine tools like Xcode on the Mac.

Quality checks

Useful guardrails before code reaches a commit.

✓

Swift hygiene checks for print, try!, fatalError, TODO, and FIXME.

✓

Design-system checks for hardcoded colors, raw RGB/hex values, and system font usage.

✓

File metadata checks for unwanted generated headers and project noise.

✓

Localization support around xcstrings read/write and localization audits.

✓

Kotlin hygiene checks for println, !! null assertions, and TODO/FIXME comments.

✓

Structured build result objects that keep errors compact enough for an AI agent to act on.

Expansion plan

More tools, more platforms.

The project should grow with real development needs. When a repeated task becomes important enough in daily work, it can become a safe MCP tool with clear inputs, bounded output, and structured results.

Web apps

  • Build orchestration for Vite, React, and Next.js
  • Preview server management
  • Playwright or Cypress tests
  • Lighthouse and bundle-size checks

Backend services

  • REST and GraphQL endpoint tests
  • Health-check suites
  • Docker image builds
  • Docker Compose stack control

Release workflows

  • TestFlight and Play Store upload helpers
  • Version and build-number checks
  • Rollback support
  • Environment-aware deployment steps

Quality and observability

  • Coverage summaries
  • Security and dependency scans
  • Crash and log analysis
  • Performance regression checks

Why it matters

AI help is better when it can verify its work.

Without tool access, an AI assistant can suggest code but cannot reliably prove the project builds, the simulator is alive, the app launches, or a lint/test suite passes.

DevTools MCP turns those checks into callable, repeatable operations. That keeps the agent useful while keeping the developer in control of the local machine and project configuration.

View GitHub Future writeups