exposing-agentkit-via-mcp

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

• Third-party content exposure detected (medium risk: 0.30). The workflow passes the user’s free-text message directly into a LangGraph ReAct agent (agent.ainvoke({"messages": ...})), and then that agent invokes MCP-discovered tools that can fetch emails and create calendar events, so outsider-authored text reaches the runtime LLM without needing selection of a specific item first.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill loads tool definitions at runtime from an MCP server URL (mcp_url) via MultiServerMCPClient.get_tools(), and those remote tool schemas (e.g., https://mcp.scalekit.com) directly control the agent's available tools and therefore its prompts/behavior.