OpenAI Apps MCP

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill provides boilerplate code and documentation for creating MCP servers. It uses standard industry libraries such as @modelcontextprotocol/sdk, hono, and zod from official registries.
  • [SAFE]: All external documentation and repository links point to official project sites (OpenAI, Cloudflare, Model Context Protocol) or well-known community repositories.
  • [INDIRECT_PROMPT_INJECTION]: The provided code templates define tool handlers that ingest user-provided arguments. While the examples demonstrate basic functionality like greeting a user or displaying a map, they do not include explicit input sanitization or safety boundary markers. Users building on these templates should implement validation and sanitization when processing untrusted data.
  • Ingestion points: src/index.ts via CallToolRequestSchema processing request.params.arguments.
  • Boundary markers: None provided in the simplified code examples.
  • Capability inventory: The MCP server generates tool responses and passes data to interactive HTML widgets via _meta.initialData.
  • Sanitization: Absent in the simplified demonstrative examples (e.g., direct interpolation of data.name into HTML).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 06:55 AM
Security Audit — agent-trust-hub — OpenAI Apps MCP