我为 Swift、Xcode 和 iOS Simulator 构建 MCP 工具箱

AI coding agents are getting surprisingly good at writing Swift. But generating Swift code is only a small part of actually building an iOS app. A real iOS development workflow involves much more. Can an AI agent understand the structure of an existing Swift project? Can it review SwiftUI architecture? Can it catch Swift concurrency problems? Can it build the project with Xcode? Can it boot an iOS Simulator? Can it install and launch the app? Can it capture screenshots, inspect logs, understand a failure, make a fix and test everything again? That is the problem I’m exploring with ios-agent-mcp. What is ios-agent-mcp? ios-agent-mcp is an open source MCP server built specifically for iOS development and AI coding agents. Instead of only giving an agent access to source code, the idea is to expose more of the actual iOS development environment. The project currently covers or is expanding toward: • Swift project analysis • SwiftUI review • Swift concurrency review • architecture analysis • API availability checks • App Store readiness auditing • Xcode project inspection • xcodebuild • xcrun • simctl • iOS Simulator discovery • Simulator boot and shutdown • app installation • app launching • screenshot capture • Simulator and application logs • device discovery • build workflows • test workflows • debugging workflows • MCP integration with coding agents Why build this? Most AI coding workflows still look something like this: prompt → generate code → developer runs it → error → copy error back to AI → generate another fix I want to reduce that manual loop. A coding agent should eventually be able to work more like this: project inspection → code analysis → Xcode build → Simulator → launch → observe → logs → debug → fix → test again The important part is the feedback loop. Writing code should not be where the agent stops. Swift and SwiftUI understanding An iOS specific agent should understand more than syntax. It should be able to inspect how a project is structured and reason about things such as: SwiftUI state management navigation view architecture async and await actors Sendable issues MainActor usage API availability deployment targets project configuration potential App Store problems The goal is to give the agent tools designed specifically around these problems rather than expecting a general purpose model to infer everything from source files. Xcode tooling Another major part of the project is connecting agents to the tools iOS developers already use. That includes: xcodebuild xcrun simctl These tools make much more interesting agent workflows possible. For example, an agent could build a project, inspect the build output, identify the error, modify the relevant code and build again. That starts moving from simple code generation toward an actual development loop. iOS Simulator Simulator automation is one of the areas I’m especially interested in. An agent should eventually be able to: discover installed Simulator runtimes list available devices boot a device shut down a device install an application launch an application capture screenshots read application logs inspect runtime failures run tests This opens up some interesting possibilities. Imagine giving an agent a task like: “Build this screen and make sure it works on the latest iPhone Simulator.” Instead of only returning SwiftUI code, the agent could actually build the project, boot the Simulator, launch the app and inspect the result. The bigger idea I don’t want ios-agent-mcp to become just another collection of prompts. The goal is to build a toolbox that lets coding agents interact with the real iOS development environment. Eventually I want the workflow to feel closer to: Understand → Build → Run → Observe → Debug → Fix → Verify There is still a lot to build, but I think this is one of the more interesting directions for AI assisted iOS development. Open source The project is completely open source. GitHub: https://github.com/Nagarjuna2997/ios-agent-skill npm: https://www.npmjs.com/package/ios-agent-mcp You can install it through npm and follow the project on GitHub as it evolves. I’d love feedback If you work with Swift, SwiftUI, Xcode, MCP servers or AI coding agents, I’d love to hear what you think. What part of iOS development do current coding agents struggle with the most? What Xcode or Simulator capability would you want an AI agent to have? And what should I build next? swift #ios #mcp #ai