{"slug": "bypassing-adb-how-i-built-a-60hz-android-ui-daemon-in-kotlin-and-a-zero-token-a", "title": "Bypassing ADB: How I Built a 60Hz Android UI Daemon in Kotlin and a Zero-Token A* MCP Router in Rust", "summary": "An open-source project called DroidPath pairs a headless Kotlin daemon with a cross-platform Rust MCP server to give AI agents sub-second access to physical Android devices without root, streaming AccessibilityNodeInfo over raw TCP at 60Hz on port 27189 via adb forward. DroidPath's author reports that standard tools such as adb shell uiautomator dump, Appium and Maestro take 2–4 seconds per tap, while sending screenshots to a multimodal model on every tap costs $0.05–$0.10 per run; DroidPath instead caches known UI flows in a local A* spatial graph to cut token costs to zero on those routes. The Rust server adds FNV-1a screen-layout hashing with PII redaction, a macro execution engine with self-healing, and a hazardous-action classifier that gates risky steps behind human approval.", "body_md": "# Bypassing ADB: How I Built a 60Hz Android UI Daemon in Kotlin and a Zero-Token A* MCP Router in Rust\n\nHow DroidPath streams Android AccessibilityNodeInfo over raw TCP at 60Hz and caches UI flows in a local A* graph so AI agents navigate real devices in under a second. f you have ever wired an LLM agent (Cursor, Claude Code, Gemini CLI, or a local model) to a physical Android device, you run into the\n\nsame three bottlenecks almost immediately: ADB latency. Standard tools like adb shell uiautomator dump, Appium, or Maestro spawn processes and serialize XML on every step. Waiting 2–4 seconds per tap turns a 5-screen flow into a 20-second crawl. Vision-agent token burn.\n\nSending screenshots to a multimodal model on every single tap costs $0.05–$0.10 per run and adds seconds of inference lag. In CI regression suites, that bill scales fast. Espresso maintenance cost. Espresso runs fast and costs $0 in compute, but the moment a layout changes or a viewId gets refactored, an engineer burns hours fixing flaky locators.\n\nI built DroidPath — an open-source headless Kotlin daemon and cross-platform Rust MCP server — to give AI agents sub-second access to physical Android devices without root, while driving token costs on known routes down to zero. Android lets you execute compiled DEX code directly in the system environment under the shell user via app_process — no root, no unlocked bootloader, no custom ROM. DroidPath splits the workload into two layers: [ AI Agent (Cursor / Claude / Gemini CLI / Local LLM) ] │ │ MCP (stdio / local HTTP+SSE) ▼ [ DroidPath Desktop & MCP Server (Rust + Tauri v2) ] ├── ScreenLayoutHasher (FNV-1a hashing + PII redaction) ├── SpatialGraphEngine (Local DAG + A* pathfinding) ├── MacroExecutionEngine (Execution + self-healing) └── HazardousActionClassifier (Human-in-the-Loop gate) │ │ Raw TCP 60Hz (127.0.0.1 via adb forward) ▼ [ Android Headless Daemon (Kotlin, app_process) ] └── AccessibilityNodeInfo -> Shadow DOM -> TCP :27189\n\n## Key Takeaways\n\n- •Standard tools like adb shell uiautomator dump, Appium, or Maestro spawn processes and serialize XML on every step.\n- •This story was reported by **Dev.to** , covering developments in the**dev** space.\n- •AI advancements continue to reshape industries — read the full article on Dev.to for complete coverage.\n\n📖 Continue reading the full article:\n\n[Read Full Article on Dev.to →](https://dev.to/v0val0x/bypassing-adb-how-i-built-a-60hz-android-ui-daemon-in-kotlin-and-a-zero-token-a-mcp-router-in-rust-8h6)", "url": "https://wpnews.pro/news/bypassing-adb-how-i-built-a-60hz-android-ui-daemon-in-kotlin-and-a-zero-token-a", "canonical_source": "https://ainexusdaily.vercel.app/article/2026-11-22-bypassing-adb-how-i-built-a-60hz-android-ui-daemon-in-kotlin-and-a-zero-token-a-", "published_at": "2026-11-22 13:51:47+00:00", "updated_at": "2026-11-22 14:13:41.388888+00:00", "lang": "en", "topics": ["ai-agents", "agent-protocols", "ai-tools", "developer-tools"], "entities": ["DroidPath", "Kotlin", "Rust", "MCP", "ADB", "Appium", "Maestro", "Dev.to"], "also_reported_by": [], "takeaway": "Replacing per-step ADB uiautomator dumps and screenshot-based vision calls with a 60Hz AccessibilityNodeInfo TCP stream plus a cached A* UI graph cuts agent navigation to under a second and drops token cost to zero on known routes, versus $0.05–$0.10 per run for screenshot-driven agents.", "recommendation": {"advice": "Cache known UI flows in a local A* graph and stream AccessibilityNodeInfo over TCP instead of dumping XML or sending screenshots each step to cut per-tap latency and token spend.", "playbook": "agent-architecture", "playbook_url": "https://wpnews.pro/playbooks/agent-architecture"}, "alternates": {"html": "https://wpnews.pro/news/bypassing-adb-how-i-built-a-60hz-android-ui-daemon-in-kotlin-and-a-zero-token-a", "markdown": "https://wpnews.pro/news/bypassing-adb-how-i-built-a-60hz-android-ui-daemon-in-kotlin-and-a-zero-token-a.md", "text": "https://wpnews.pro/news/bypassing-adb-how-i-built-a-60hz-android-ui-daemon-in-kotlin-and-a-zero-token-a.txt", "jsonld": "https://wpnews.pro/news/bypassing-adb-how-i-built-a-60hz-android-ui-daemon-in-kotlin-and-a-zero-token-a.jsonld"}}