Clawperator

Deterministic Android Automation for AI Agents

Let AI agents use and control Android apps on behalf of users.
Clawperator connects your agent to a dedicated Android device so it can observe the screen, perform UI actions, and receive structured results.

Quick Start

# Install Clawperator on macOS/Linux

curl -fsSL https://clawperator.com/install.sh | bash

Features

Tools agents can trust: deterministic execution, structured outputs, and composable automation.

Deterministic

Strict contracts, no hidden retries, and a single canonical result envelope for every command.

Observable

Rich, structured UI snapshots and explicit terminal markers for reliable agent state tracking.

Agent-First

Machine-readable error codes and JSON-native output designed for robust branching logic.

Why Clawperator?

Many consumer services only expose critical data in mobile apps, not public web APIs.

Family tracking and location apps
Home automation apps
Grocery and shopping apps
Ride-hailing apps

Clawperator lets AI agents interact with these apps on your behalf. The interface stays predictable and stable so skills can be created, reused, and shared.

What Is Clawperator?

Clawperator is the execution layer for LLM-driven Android automation. It provides a deterministic Node.js CLI and HTTP API, the hand for an LLM brain.

What agents can do

Connect to a real Android device
Run deterministic UI actions (tap, scroll, type, read)
Observe screen state via structured snapshots
Get canonical terminal results (`[Clawperator-Result]`)
Compose primitives into repeatable skills

Design principles

Deterministic: Strict contracts, no hidden retries, one result envelope per command.
Observable: Structured UI snapshots and machine-readable error codes.
Agent-first: JSON output, typed errors, single-flight concurrency.

Typical setup: A dedicated Android device (any cheap or old phone) stays connected to your host machine as a permanent actuator. Your agent sends commands through Clawperator, and Clawperator executes and reports results.

How It Works

The hand loop is simple and deterministic: observe, decide, execute, and report.

Observe

Capture structured UI state from the Android device so the agent can reason on a stable snapshot.

Decide

The brain chooses the next action based on selectors and current state. Clawperator does not plan.

Execute

Run deterministic device actions (tap, type, scroll, read) with strict validation and no hidden retries.

Return

Emit one canonical terminal envelope (`[Clawperator-Result]`) with machine-readable success or failure.