Your AI agent doesn't need screenshots to read text. ScreenRead gives it the macOS accessibility tree in ~100ms. CLI + MCP server.
Bambushu / screenread
