Apple announced Xcode 26.3 on Tuesday, introducing agentic coding tools such as Anthropic’s Claude Agent and OpenAI’s Codex directly within its app development suite. The Xcode 26.3 Release Candidate became available to all Apple Developers on the developer website the same day, with a full App Store release to follow shortly.
This update builds on Xcode 26, released the previous year, which added support for ChatGPT and Claude inside the integrated development environment (IDE) used to build apps for iPhone, iPad, Mac, Apple Watch, and other Apple platforms. The new agentic coding features enable AI models to access more Xcode functionalities for complex automation and project management tasks. These models also connect to Apple’s current developer documentation, ensuring they incorporate the latest APIs and best-practice guidelines.
At launch, the agents assist developers by exploring project structures and metadata, building projects, running tests to identify errors, and fixing issues automatically. Apple collaborated with Anthropic and OpenAI to refine the integration, with specific efforts to optimize token usage and tool-calling efficiency.
Xcode uses the Model Context Protocol (MCP) to expose its capabilities to agents and link them with tools including project discovery, file management, previews, snippets, and documentation access. This MCP compatibility allows Xcode to work with any external MCP-supported agent.
To use the features, developers download desired agents from Xcode’s settings and connect their AI provider accounts via sign-in or API keys. A drop-down menu lets them select specific model versions, such as GPT-5.2-Codex or GPT-5.1 mini. Developers then enter natural-language commands in a prompt box on the left side of the screen, instructing the agent on tasks like adding features using Apple frameworks, including how they should appear and function.
Once prompted, the agent breaks tasks into smaller steps and consults relevant documentation before coding. It visually highlights code changes and displays a project transcript on the side of the screen, revealing its internal processes. Apple states this transparency benefits new developers learning to code.
Apple scheduled a “code-along” workshop for Thursday on its developer site. Participants can watch demonstrations and practice agentic coding tools in real time using their own Xcode installations.
After completing tasks, the agent runs tests to verify code functionality. Based on test results, it iterates to resolve errors or other issues. Apple recommends prompting the agent to plan ahead before coding, as this pre-planning step improves outcomes. Developers retain control, with Xcode automatically creating milestones for every agent-made change, allowing easy reversion to prior code states at any time.
The agentic tools extend Xcode’s AI capabilities beyond basic chat integrations in Xcode 26. Agents now perform end-to-end workflows, from project analysis to testing and refinement. Developers issue commands like directing the addition of app features tied to specific Apple frameworks.
Setup remains straightforward: agents install via settings, followed by account linkage. The interface includes the prompt box for instructions, task decomposition visuals, code highlights, and transcripts for oversight.
Access to live documentation ensures agents reference up-to-date resources. MCP underpins tool interactions, supporting broad agent compatibility.
The Release Candidate availability marks immediate testing for developers, ahead of the App Store version. This positions Xcode as a hub for advanced AI-assisted development on Apple platforms.
