vercel-ai-sdk

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill includes Model Context Protocol (MCP) HTTP/SSE endpoints (e.g., "https://mcp-server.example.com/mcp" and "https://mcp-server.example.com/mcp/sse") that are fetched at runtime to load tools/prompts which directly influence agent behavior, and it also shows Stdio transports that run "npx" to fetch+execute remote MCP server packages at runtime.