- Home
- Skills
- Mobile Development
- ADB Pilot: Android Control for MCP Agents
Works with the AI Agents you already use
ADB Pilot: Android Control for MCP Agents
Give your AI agent safe, structured control of Android devices and emulators over MCP — screenshots, logcat, installs, and UI taps behind approval gates.
Free
ADB Pilot: Android Control for MCP Agents
Example session with this skill installed
I'm building a mobile app and I want my agent to verify it on the emulator instead of me typing adb commands all day. Repo is at ~/projects/mobile-app, the debug build is /tmp/app-debug.apk, and emulator-5554 is running. Help me configure an ADB Pilot MCP server for Cursor using stdio transport, show which safety layers to enable before I expose shell access, then walk me through a crash-triage run: start com.acme.mobile, capture ten seconds of logcat, and summarize the failure.
- Read your context and instructions
- Compiled the adb pilot
Summary
ADB Pilot is now wired into Cursor over stdio with the approval-gating policy enabled. The crash-triage run on emulator-5554 reproduced the failure and pinpointed a null cart session in PaymentFragment.onResume().
| Check | Before | After |
|---|---|---|
| ADB access | manual adb typing | adb-pilot --transport stdio in mcp.json |
| Device targeting | guessed device names | serial emulator-5554 required per call |
| Risk gating | none | reboot / uninstall / clear-data approval-gated |
| Crash cause | unknown | NPE at PaymentFragment.kt:87, logged 14:32:07 |
FATAL EXCEPTION: main
java.lang.NullPointerException: getSession(...) must not be null
at com.acme.mobile.PaymentFragment.onResume(PaymentFragment.kt:87)
Next steps
- Guard
CartRepository.sessioninonResumewith a recreate-or-return path, then reinstall/tmp/app-debug.apk - Add the default denylist (
/system,/vendor,su, rawpmwrites) before enabling the shell tool - Re-run the UI smoke-test recipe to confirm the payment screen renders after the fix
- Keep one clean emulator profile reserved for regression runs
Connects securely to your tools. The creator never sees your data.
About this skill
Controlling Android devices via AI agents often results in flaky execution or dangerous shell access. ADB Pilot provides a structured bridge using the Model Context Protocol (MCP) to give agents safe, granular control over physical devices and emulators. It replaces raw terminal commands with typed tools for UI automation, diagnostic retrieval, and app management.
What it does
- Device Management discovers connected hardware, identifies serials, and monitors connection health.
- UI Automation executes precise taps, swipes, and text input using coordinate or hierarchy-based targeting.
- Diagnostic Capture retrieves logcat streams, crash reports, ANR files, and memory dumps for triage.
- App Operations handles APK installation, permission grants, and application lifecycle management.
- File Transfer manages file round-trips between the host machine and device storage paths.
- Visual Verification captures screenshots and screen recordings to verify agent actions.
How it works
- Bridge Setup involves connecting a device via USB or TCP/IP and starting the MCP server on the host.
- Tool Invocation allows the agent to call specific functions like
install_apkortap_uithrough the MCP client. - Safety Gating enforces a risk-based approval model, requiring human confirmation for destructive actions like reboots or deletions.
- Result Normalization converts raw ADB output into structured JSON for the agent to process.
Frameworks & tools
This skill utilizes the Model Context Protocol (MCP), Android Debug Bridge (ADB), and Python. It supports local execution via stdio or remote access through SSE/HTTP transports.
Why this beats prompting it yourself
Writing raw ADB shell commands through an LLM is error-prone and lacks safety boundaries. This skill provides a schema-validated interface that prevents shell injection, enforces device targeting, and categorizes operations by risk level.
Use cases
- Automate UI smoke tests on physical devices during CI/CD cycles.
- Triage application crashes by automatically pulling logs and screen states.
- Manage large emulator fleets for parallel app testing and data collection.
- Perform remote device surveys to verify hardware specs and battery health.
Known limitations
Requires ADB to be pre-installed and available on the system PATH. Containerized deployment requires complex host USB mapping. Destructive operations on system partitions are blocked by default.
It is the bridge design and operations blueprint, not a prebuilt server — you supply or install an ADB-over-MCP implementation yourself. Docker-mode ADB still needs host USB access, which is fiddly on some platforms.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
1 use
Times used to date
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 1 month ago
- Free with an Agensi account
Requires
Permissions
File Scopes
Shell access runs scoped ADB commands against devices you own; file read/write covers MCP client configs, logs, and screenshot artifacts, with no network exposure required.