Claude Code in Xcode: Claude Agent, MCP and Which Setup to Use

Yes, and Apple has built it in. Xcode's Intelligence settings have an Agents section where Claude Agent installs with a click, and a Model Context Protocol section that lets agents running outside Xcode, Claude Code included, use Xcode's tools. Apple's own documentation gives the exact Claude Code command for the second route. So the real question is not whether it works, it is which of the two setups fits how you work.

This is the fifth entry in my editor series, after the Claude Code VS Code extension, the Claude Code JetBrains plugin, Claude Code in Zed and Claude Code in Cursor. Xcode is different from all of those. There is no extension to install from a marketplace. Apple ships the hooks, and you choose whether Claude lives inside Xcode or talks to it from outside.

Per Apple's docs, an agent you choose in the coding assistant automatically has access to Xcode capabilities such as building and testing your app. That is the big difference from pasting code into a chat window: Claude can make a change, build, see what broke and fix it, in the same place you would.

Xcode keeps agent downloads up to date automatically where it can. If Claude Agent is not listed where you expect, Apple's docs point to the Components area of Xcode's settings for managing agent downloads.

A few things inside Xcode will feel familiar if you have used Claude Code anywhere else. Type / in the message field to see the skills Xcode offers. Use /plan to have the agent explore and propose an approach without touching your code, and /exit-plan to leave plan mode. Apple's docs say Xcode also enters plan mode on its own when your prompt asks for a plan. I start every non-trivial change this way, for the same reasons I cover in Claude Code plan mode.

This is the command straight from Apple's documentation. xcrun mcpbridge is the bridge Xcode provides, and the claude mcp add part registers it with Claude Code under the name xcode. Xcode shows an alert when an external agent connects and while it is active, so you will always know when Claude is working through it.

Two things trip people up. First, Claude Code adds MCP servers to the current project by default, so run the command from your project folder, or add --scope user if you want the Xcode tools available everywhere. Second, the project has to be open in Xcode before you prompt, because the bridge works through the open project. If Claude says it cannot see the Xcode tools, Claude Code MCP server not working walks through the fixes.

Apple also suggests adding hints about Xcode and your project to your CLAUDE.md. I would do it: a few lines on your scheme name, how you run tests and which simulator to use save Claude from guessing every session. The CLAUDE.md file covers what to put in it.

My take: Route 1 is the easiest start for a native Swift app, because everything happens in Xcode with no terminal and the agent can build and test from day one. Route 2 is the better long-term setup if Claude Code is already part of how you work, because one setup follows you across every project instead of keeping a second copy just for Xcode. You can run both. Just do not point two agents at the same files at the same time.

If you are not tied to Swift, how to build a mobile app with Claude Code shows the Expo route, and the Claude Code desktop app is where I tell most people in the club to start. Claude Code day one setup covers the first hour either way.