mcporter

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). The workflow’s mcporter list/mcporter call can target ad-hoc MCP endpoints (e.g., mcporter list https://...), and at runtime MCPorter will fetch/discover server-provided tool metadata and tool docs from that outsider-controlled endpoint, which then gets rendered as readable text in the agent/LLM context (indirect prompt injection risk via server-supplied descriptions/examples).

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

• Potentially malicious external URL detected (high risk: 0.90). Yes — the skill shows mcporter calling ad-hoc MCP endpoints at runtime (e.g., https://mcp.linear.app/mcp and https://www.shadcn.io/api/mcp.getComponent(...)), which are fetched live to obtain tool schemas/signatures that directly control the agent's prompts/invocations.