ForrestKim42/llm-app-exploration

View on GitHub ↗

A pattern for LLM agents to explore and control any app — mobile or desktop. Accessibility-first, not vision-based. Share with your agent and go.

23 ★4 forksUpdated 5mo ago

What you need to know

An idea/methodology document describing a pattern for systematically exploring any app (mobile or desktop) with an LLM agent, producing a route map (ROUTES.md), transition table (TRANSITIONS.md) and user flows. References the companion mobile-mcp and desktop-mcp servers.

Key features

  • ✓Graph-based app exploration using the accessibility tree (DFS)
  • ✓Stable TYPE:Label element identity
  • ✓ROUTES.md, TRANSITIONS.md and parameterized flows artifacts
  • ✓Depth levels L1 screens, L2 interactions, L3 real actions

Best for

Teams wanting LLM agents to systematically explore and document app UIs

Caveats

  • ⚠This repo is a pattern description, not a standalone MCP server implementation
  • ⚠Two companion MCP servers (mobile-mcp, desktop-mcp) implement the pattern

Reviewed 2026-08-11

Topics

adbandroidapp-explorationautomationclaude-codellmmcpmobile-testingmodel-context-protocolux-research
Stars
23★
Forks
4
Language
—
License
MIT
Created
2026-04-07
Last push
2026-04-08