Xcode AI Tooling Advances

An iPhone app used to be built with two separate layers: code that makes screens, and a second layer that teaches Siri, Spotlight, and Shortcuts what the app can do. Apple’s latest App Intents updates and the rise of tools like XcodeBuildMCP are pulling those layers closer together. (developer.apple.com) (github.com) App Intents are the labels and switches that describe an app’s actions in a machine-readable way. Apple says they make an app’s content and actions discoverable across Siri, Spotlight, widgets, controls, and the Shortcuts app. (developer.apple.com) That means a developer is no longer only building a button inside the app. The same action can also show up as a voice request in Siri, a result in Spotlight search, or a reusable step inside a Shortcuts automation. (developer.apple.com 1) (developer.apple.com 2) Apple pushed that system further at Worldwide Developers Conference 2025 with new App Intents features like interactive snippets, entity view annotations, deferred properties, and new integrations for Spotlight and Visual Intelligence. Apple described the goal as making App Intents “more expressive” while also becoming easier to adopt. (developer.apple.com) Apple is also tying App Intents more directly to Apple Intelligence. Its documentation now tells developers to create intents, entities, and enumerations that conform to assistant schemas so Siri and Apple Intelligence can use an app’s actions with stronger action-taking capabilities. (developer.apple.com 1) (developer.apple.com 2) On a separate track, the Model Context Protocol is becoming a standard way to let an artificial intelligence coding assistant call tools instead of only writing text. XcodeBuildMCP is one of the clearest examples in Apple development, because it exposes Xcode project, simulator, device, and Swift Package Manager operations through a tool interface an agent can invoke directly. (github.com) (developer.apple.com) In plain English, that turns “please fix this iPhone build error” from a chat prompt into a loop. The assistant can inspect the project, run a build, launch a simulator, collect logs, and try again without the developer manually typing every terminal command. (github.com) The current XcodeBuildMCP project is not a tiny experiment anymore. The main repository shows more than 5,000 GitHub stars, a recent v2.3.1 release, and setup paths for Homebrew, npm, Cursor, and OpenAI Codex command-line interface use. (github.com 1) (github.com 2) Apple’s own platform direction lines up with that shift. The new Foundation Models framework lets developers use the on-device model behind Apple Intelligence, and Apple explicitly says tool calling lets that model call code to fetch current information or perform actions. (developer.apple.com) (developer.apple.com) Put those pieces together and the workflow changes shape. App Intents give apps a structured map of what actions exist, and tool-calling systems like XcodeBuildMCP give coding agents a structured way to build, run, and debug the app that contains those actions. (developer.apple.com) (github.com) The practical consequence for teams is not that artificial intelligence suddenly writes the whole app. It is that more of the repetitive wiring work — scaffolding a project, exposing actions, running builds, checking logs, and iterating on integration code — is moving from ad hoc terminal commands into machine-callable interfaces. (developer.apple.com) (github.com) That usually speeds up the first draft and raises the cost of weak guardrails. If an agent can modify intent definitions and also trigger builds automatically, teams will need tighter review around generated code, permissions, simulator or device actions, and the exact steps allowed inside continuous integration pipelines. (github.com) (developer.apple.com)