mcp-apps-protocol

Pass

Audited by Gen Agent Trust Hub on Aug 13, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a protocol for processing external data within a sandboxed UI environment. It correctly identifies the attack surface where untrusted structuredContent from a server is rendered in the View.
  • Ingestion points: The ui/notifications/tool-result method delivers server-provided content to the View (SKILL.md).
  • Boundary markers: The documentation explicitly instructs developers to treat all incoming content as untrusted.
  • Capability inventory: The View is restricted to a specific JSON-RPC bridge for host communication (tools/call, ui/message, ui/update-model-context).
  • Sanitization: The specification mandates rendering content as text and explicitly prohibits the use of innerHTML or eval to prevent code execution vulnerabilities.
  • [SAFE]: The skill is purely informational/reference-based and does not contain any executable scripts, binary files, or installation commands. It promotes secure development practices such as restrictive CSP headers and iframe sandboxing.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 13, 2026, 10:14 AM
Security Audit — agent-trust-hub — mcp-apps-protocol