C — Moderate risk signals Graded October 16, 2026Several concerns worth a closer look before you install.Based on public adoption and transparency signals as of October 16, 2026. Our opinion, not a code audit or security guarantee.
App connector · Developer tools, Web & browsing
agent-device
Mobile app automation and verification for AI coding agents. CLI, MCP server, and typed Node.js API for iOS, Android, HarmonyOS, TV, web, macOS, and Linux.
Bottom line: Signal-based grade (2/7 points): heavily downloaded, no public source code found. This reflects adoption and transparency, not a code review.
- Popularity score
- 41/100
- Rank
- #294 of 500
- Weekly downloads
- 296k
What it does
agent-device enables AI coding agents to automate and verify interactions across mobile apps (iOS, Android, HarmonyOS, TV), web, macOS, and Linux through a CLI, MCP server, or typed Node.js API. It lets agents programmatically control devices, capture screenshots, and validate app behavior—useful for testing, QA automation, and agent-driven mobile workflows.
Who it’s for
AI builders and engineering teams automating mobile app testing, QA verification, or device interaction workflows. Enterprises running AI agents that need to interact with or validate behavior across multiple device platforms.
Copy-paste setup
Check these steps against the project’s own instructions before running anything.
Install via npm:
npm install agent-device
For Claude Desktop MCP integration, add to claude_desktop_config.json:
{
"mcpServers": {
"agent-device": {
"command": "npx",
"args": ["agent-device"]
}
}
}
Verify against the repo README for environment setup (device authentication, platform-specific config) and any required dependencies.
Risks
- Device credential scope: Requires API keys or tokens to control real devices; misconfigured permissions could allow unauthorized device access or app manipulation.
- Prompt injection via UI automation: Agents interpreting screen text or app responses as instructions could be exploited if app content is untrusted or adversarially crafted.
- Data exfiltration: Agents capturing screenshots or device state may inadvertently extract sensitive user data (passwords, PII, auth tokens) from app screens.
- Supply-chain risk: High weekly download volume (295k) means vulnerabilities or compromised updates could affect many downstream users and agent deployments.
- Minimal public scrutiny: 0 GitHub stars and 0 HN points suggest limited security vetting or community review compared to established alternatives.
How to stay secure
agent-device Security Hardening Guide
Least-privilege permissions
- Limit device access credentials: Issue separate API keys for each AI agent and device, so if one key is compromised, only that device is affected—not your entire fleet.
- Restrict app permissions: Configure the connector to control only the specific apps and features your agents need; don't grant blanket access to all installed applications.
- Rotate credentials regularly: Change API keys and tokens every 30–90 days, especially after staff changes or suspected exposure.
- Disable unused features: Turn off screenshot capture, clipboard access, or device state logging if your agents don't need them; fewer capabilities = fewer ways to leak sensitive data.
Network isolation
- Run on private networks: Deploy the MCP server on internal networks only; never expose it directly to the internet without a VPN or firewall.
- Segment agent environments: Isolate devices and agents by purpose (e.g., testing, production); use separate network zones so a compromised test agent can't reach production devices.
- Encrypt all communication: Enforce TLS/SSL encryption (HTTPS) between agents, the connector, and devices to prevent credential or data interception.
- Audit external dependencies: Verify that the connector doesn't phone home or communicate with untrusted external services; block outbound traffic except to known, approved destinations.
Monitoring
- Log all device interactions: Record which agent accessed which device, what actions were performed, and when—review logs weekly for suspicious patterns.
- Alert on anomalies: Set up alerts for unusual activity (e.g., large screenshot volumes, credential reuse across different agents, or access outside normal hours).
- Track supply-chain updates: Monitor for security patches and changelogs from the vendor; apply updates within 7 days of release, given the high download volume and limited public vetting.
- Audit data captures: Periodically review stored screenshots and device logs for inadvertently captured sensitive information (passwords, tokens, PII); delete or redact as needed.
Red flags to watch for
- Unvetted app content: If your agents are testing or automating untrusted or user-generated apps, assume the app UI or responses could contain malicious instructions designed to fool the agent.
- Overpermissioned keys: API keys with access to all devices or all features; keys shared across teams or stored in code repositories; credentials that haven't been rotated in over 90 days.
- Unexpected data volumes: Unusually high screenshot capture rates, large data transfers, or new outbound connections could signal exfiltration or a compromised agent.
- Infrequent updates or quiet issues: Long gaps between vendor releases, unresponded security reports, or lack of public discussion (0 GitHub stars, no community activity) may indicate limited ongoing maintenance or scrutiny.
Alternatives
- Appium – Open-source mobile automation framework; mature, widely audited, but requires separate MCP wrapper or custom integration.
- BrowserStack/Sauce Labs – Commercial cloud device labs with built-in security controls; higher cost but managed infrastructure and compliance.
- Maestro – Modern mobile automation tool with simpler syntax; smaller ecosystem than Appium but growing adoption.
- Native platform SDKs (XCTest, Espresso, UIAutomator) – Platform-specific and labor-intensive but offer fine-grained control and no third-party trust dependency.