Scored across 77 tools
Most tools have clearly distinct purposes and descriptions explicitly guide when to use each (e.g. describe vs native-describe-screen vs debugger-component-tree, gesture-swipe for iOS/Android vs gesture-scroll for Chromium, view-network-logs vs native-network-logs). However, the discovery/inspection layer has real overlap — five tree/hierarchy readers plus three describe-family tools — and the seven gesture tools (tap/swipe/scroll/drag/custom/pinch/rotate) require reading platform notes to pick correctly, creating occasional misselection risk.
Agents benefit from consistent prefix namespacing across families (gesture-*, flow-*, debugger-*, native-*, react-profiler-*, profiler-*, chromium-*), which makes the set navigable. Still, the verb/noun ordering drifts (list-devices/launch-app vs boot-device/screen-recording-start vs stop-all-simulator-servers) and several bare single-word tools (button, keyboard, text, key, describe, shake, fold) break the pattern.
At 77 tools this is far above any reasonable single-server surface, dwarfing the 3–15 sweet spot and even the 25+ 'too many' threshold. The domain genuinely spans several subsystems (device control, recording, network, JS/native debuggers, three profiler suites, flow authoring, self-update), but the profiler and query variants alone (react-profiler-*, native-profiler-*, profiler-cpu/commit/stack-query, profiler-load/combined-report) constitute an entire sub-suite that bloats the surface well past what the scope justifies.
Coverage is unusually thorough: boot/launch/restart/reinstall app lifecycle, permissions, full gesture and input surface, screenshot/recording/diff capture, accessibility and native view inspection, JS and native network capture, debugger and two profiler stacks with analysis and drill-down, flow record/replay, and explicit teardown plus self-update tools. Few obvious dead ends remain for the stated device-automation purpose.