Repository navigation
Replies: 1 comment 2 replies
|
I see my agents doing similar, but don't see much of a difference between a |
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Experimenting Maestro MCP for coding-agent validation on Android/iOS.
One pattern we're seeing is very incremental exploration:
inspect -> tap -> inspect -> type -> inspectI saw that 2.5.0 intentionally removed action-specific MCP tools in favor of
run, which makes sense to avoid mirroring the whole Maestro DSL as MCP tools.I'm wondering if there's still room for a single imperative primitive such as
interact, specifically for exploratory agent interaction, while keepingrunfor reproducible flows.Something like:
interact(action="tap", id="login_button", args...)The distinction I'm thinking about is:
Is keeping
runas the only interaction primitive an intentional long-term direction, or would something like this be worth experimenting with?All reactions