building-mcp-servers
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill facilitates fetching official protocol documentation and SDK README files from trusted domains such as
modelcontextprotocol.ioandgithub.com/modelcontextprotocol. - [INDIRECT_PROMPT_INJECTION]: As a guide for building API-connected tools, the skill describes systems that ingest untrusted data from external services and LLM prompts.
- Ingestion points: MCP server tools described in the guides accept natural language parameters and process JSON/Markdown responses from external APIs.
- Boundary markers: The documentation mandates the use of strict runtime validation schemas (Zod for Node.js, Pydantic for Python, and Data Annotations for .NET) to define tool boundaries.
- Capability inventory: The guide covers tools capable of network operations and data transformation; the provided evaluation instructions involve executing these servers for automated testing.
- Sanitization: The guides provide specific implementations for input sanitization, scope enforcement in OAuth middleware, and truncation of large responses to maintain safety.
Audit Metadata