ios-mcp-jailbreak-automation

Warn

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The run_command tool provides a gateway for the AI agent to execute arbitrary shell commands on the jailbroken iOS device. This enables low-level system access and control over the device's operating system.
  • [PRIVILEGE_ESCALATION]: The documentation identifies an mcp-root helper utility that facilitates root escalation for certain operations. This allows the AI agent to perform tasks with the highest possible system permissions on the iPhone.
  • [DATA_EXFILTRATION]: Several tools are designed to extract sensitive information from the mobile device:
  • screenshot captures the current visual state of the device as a Base64 image.
  • get_clipboard accesses the system clipboard, potentially exposing passwords, codes, or private text.
  • get_ui_elements reads the accessibility hierarchy, which includes all text, labels, and structured data visible in the foreground application.
  • [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It ingests untrusted data from external sources (such as app UI content via get_ui_elements or clipboard data via get_clipboard) and possesses high-privilege capabilities like run_command and install_app. A malicious application or website could present instructions that an agent might inadvertently follow, leading to unauthorized device actions.
  • Ingestion points: get_ui_elements (UI text), get_clipboard (user data), run_command (shell output).
  • Boundary markers: None present in the instructions to separate device content from agent instructions.
  • Capability inventory: run_command (shell), install_app, launch_app, tap_screen, set_clipboard, open_url.
  • Sanitization: None detected; the agent is instructed to use the tool outputs directly for automation logic.
  • [EXTERNAL_DOWNLOADS]: The configuration requires the use of npx to fetch and run the @modelcontextprotocol/server-fetch tool. This package originates from the official Model Context Protocol organization and serves as the bridge between the agent and the mobile device.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 16, 2026, 10:27 PM
Security Audit — agent-trust-hub — ios-mcp-jailbreak-automation