How is the performance so far ?

[rams-kroger]

Hi Team, I am looking forward to use this MCP. If anybody already using so far, how is the performance and consistency of automated execution ?

[gmegidish]

@rams-kroger Great question, I'll split it into 2 parts:

1. Performance -- the mcp itself (and tools involved like mobilecli, webdriveragent, devicekit or any of the others) - are quite performant. Meaning, it's a few milliseconds to do a task. Performance of the LLM is a different story, it might take Claude Code a few seconds to understand that it needs to call mobile-mcp, and then 5 seconds to decide if it needs to take a screenshot because things might have changed. These things improve all the time with new models, and for example, Gemini is much much faster than Claude Code.

2. Consistency -- that's largely depends on the prompt that you're using, the model you're using, and the expectation. For example, a GOOD prompt for consistent automation would be: on my iphone simulator, launch the fitness app, make sure you're in the Summary tab, then find the current "Step Count" for today. Report back a simple json like {"stepCount":XXX} and no other output. If you can't find it, let me know why.. A bad prompt would be "find step count" with no other context.

Give mobile-mcp a try. Regardless mobile-mcp, prompting an LLM needs exercise. It takes a few attempts to get the hang of it.