[Bug]: Build fails in XCode even though it passes using the MCP build tool
Bug Description
In VSCode using Claude Code, model will use the build_mac_proj and it reports success.
But when I go into XCode and press the play button, it says the build failed. I have to screenshot the debug log from XCode IDE to show CC the issue.
Questions
- How can it fail to build in XCode when the MCP server build tool passed the build?
- Is there a way to let CC see the error output that is the same as you get when pressing play in XCode?
Debug Output
NA
Editor/Client
VSCode
MCP Server Version
Latest
LLM
Claude Sonnet 4
MCP Configuration
{
"mcpServers": {
"XcodeBuildMCP": {
"command": "npx",
"args": [
"-y",
"xcodebuildmcp@latest"
]
}
}
}
Steps to Reproduce
- Instruct LLM to build the app and test
- Open XCode to try testing the app manually
Expected Behavior
If it fails to build in XCode, it should also fail to build using the MCP build tools. An error log should be output exactly the same as would happen by pressing play in XCode. LLM fixes issues and tries building again in a loop until it builds successfully.
Actual Behavior
LLM uses MCP build tool. Tool reports success. LLM reports to user everything tests ok and ready for testing by user.
User opens XCode and presses play and experiences build error(s).
Error Messages
i have same issue, it's working with augment code but it's not working with cc.
I'm seeing this too. Many times the build fails but xcodebuildmcp reports it as a success.
+1
I don't seem to be experiencing this issue anymore.
Maybe it's a skill issue? On my third attempt to vibe code a simple native macOS app knowing exactly zero Swift, I managed to get it working really well. I now know about terms like AGILE, MVVM, mocking, XPC, sandboxing, epics/user stories, unit tests etc.
I set up the third attempt following strict best practices for MVVM and using AGILE methodology and I never saw any discrepancies between the MCP build failures and build failures within XCode.
I think from my own experiences, the project is large and complex, and there are some false positives in the logs. I think Cloud Code is kind of ignoring them all, even if there are some serious build failures.