Getting Started
Drengr installs in under a minute. It auto-detects your device tools, so in most cases you just install and go.
1. Install Drengr
From the Official MCP Registry
MCP clients with registry support can discover and install Drengr directly — no manual config needed.
For other clients, search "Drengr" in your MCP client's server browser or use the registry identifier: dev.drengr/server
Supported: macOS (Apple Silicon & Intel), Linux (x64 & ARM64).
The shell installer downloads a prebuilt binary to /usr/local/bin and verifies its SHA256 checksum.
🐉 See it work — one command
With a device or simulator running, this auto-detects it and an AI agent drives a real app in front of you. No config, no login, free.
Standalone runs use your own model — a cloud key (drengr key set anthropic <key>) or a local vision model via Ollama. Inside an MCP client, the client's own model drives it.
2. Verify
Run the health check. It tells you exactly what's working and what needs fixing.
If ADB or simctl shows [✗], see the device tools section below. Drengr will tell you exactly what to do.
3. Connect to Your MCP Client
Drengr auto-generates the config for your MCP client.
This detects your platform and prints the JSON config you need. You can also write it directly:
After writing the config, restart your MCP client to pick up the changes.
Android Studio (Gemini agent mode)
Android Studio can't launch local MCP servers itself — it connects over HTTP. Start Drengr's HTTP endpoint and keep it running while Studio is connected:
The setup command above writes mcp.json into your Android Studio config (Settings → Tools → AI → MCP Servers shows it). Restart Studio, then type /mcpin the Gemini panel to see Drengr's tools.
4. Device Tools
You probably don't need to do anything here.
Drengr auto-detects ADB and Xcode from your shell profile, even inside MCP clients that don't inherit your PATH. If drengr doctor shows green checks, skip this section entirely.
5. LLM API Keys (Optional)
Only needed for standalone mode
If you use Drengr through an MCP client (Claude, Cursor, etc.), skip this section. The MCP client's own LLM handles all reasoning. LLM keys are only needed for drengr run, which uses its own built-in OODA agent.
Uninstall
Removes the binary, all local data (~/.drengr/), keychain entries, and your MCP client configs — and asks our server to forget this machine: its registration is erased and its usage records are permanently anonymized. A sign-in account (if you made one) is deleted separately — email [email protected].
Offline? Machine records auto-delete after 90 days of inactivity anyway — the command prints your machine ID so you can also request immediate erasure by email.
If you installed via npm, drengr uninstall cleans up data and MCP configs, then reminds you to run npm uninstall -g drengr to remove the npm wrapper.
What's Next
Once drengr doctorshows green and you've connected your MCP client, you're ready. Ask your AI agent:
The agent will call drengr_look to see the screen, then drengr_doto interact with it. That's the core loop.