Stop feeding agents raw Xcode output. Give them the simulator.
XcodeBuildMCP turns build, launch, UI inspection, logs and debugging into a tighter agent loop built on Apple’s own tools.
See why:
https://t.co/REtwWkmdtK
XcodeBuildMCP 2.6 vs competing tools on real Claude UI automation tasks:
• Create and complete reminders
• Add a new contact
• Drive a Weather app UI, toggle every option, search London, and verify precipitation
Timeouts count as incomplete runs, not slow completions.
Just released XcodeBuildMCP 2.6.2 which fixes an issue where the ui-automation tools couldn't interact with TabViews. Also fixes `xcodebuildmcp upgrade` not surfacing the latest homebrew installed version. If you're stuck on an older version for now you can do `brew update & xcodebuildmcp upgrade`.
XcodeBuildMCP v2.6 is out 🎉!
Big update for agent-driven UI automation:
- Agents drive your app's UI in far fewer steps: ~70% faster, ~68% fewer tokens, ~76% fewer tool calls on a real task.
- `batch` runs several UI actions in a single call, so flows finish in fewer round-trips.
- `wait_for_ui` waits for the screen to be ready, so automation stops failing on timing.
- `drag` adds gestures agents couldn't do before, scroll long lists, expand sheets.
- `nextSteps` lets agents act on what UI steps to do next.
`brew install xcodebuildmcp` or upgrade using `xcodebuildmcp upgrade`
https://t.co/xMn8k7eHv1
@dearmycode@PaulSolt So this is just Simulator automation which is hard, macOS is much easier as there are public APIs for that. There are already tools available to controlling macOS apps.
XcodeBuildMCP proivdes stable agentic workflows for AI agents with highly optimised responses that results in your agent using less tokens and less guessing. Yes agents can build, install etc but they have to discover the project settings, choose the correct simulator, use multiple commands to build, install and launch apps where XcodeBuildMCP provides a single tool to “run” the app reducing latency by reducing number of AI tool call round trips. The main idea is to provide a stable tool surface for the agent. One thing agents can’t do without a tool like this is ui-automation like in the demo. The agent is doing everything, all the simulator interactions are performed by the agent so it can verify its own work. Agent let debugging is easier because each build/run tool automatically starts capturing logs so the agent can tail them in a token efficient manner. Xcodebuild isn’t optimised for agentic use, XcodeBuildMCP is.
Yesterday was my first day at @OpenAI working under the amazing @romainhuet.
I've spent the better part of two years pushing the frontier of having models write code for you, shipping updates week after week for @RepoPrompt.
Transitioning from being a founder is never easy, especially when you have a community of amazing developers who invested so much in your tool.
Thankfully Romain worked hard to ensure that all those users would be taken care in the process, and if you're one such user, you should have an email waiting in your inbox with the details!
I'm very excited to be joining this talented team and work alongside everyone at OpenAI contributing to codex.