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 a runtime call to ai.mcp.get_http_tools("https://mcp.example.com/mcp", ...) which fetches external tool definitions that directly control agent behavior, and also an example ai.mcp.get_stdio_tools call that runs "npx -y @anthropic/mcp-server-filesystem" (fetching and executing remote code) — both are runtime external dependencies that can control prompts or execute code.