build-mcp-use-client

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

• Third-party content exposure detected (high risk: 0.90). This skill explicitly connects to arbitrary external MCP servers (via user-supplied urls in mcpServers, the React Dashboard addServer/newUrl flow, and npx client connect) and then lists/reads resources, prompts, and tool outputs (e.g., session.readResource, session.callTool, resources read/subscribe, prompts get), so untrusted third-party content from those servers can be ingested and influence subsequent tool calls and decisions (see the "When to use" and "Workflow" / recipes showing readResource, callTool, and elicitation URL handling).

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

• Potentially malicious external URL detected (high risk: 0.80). The skill includes runtime fetches that directly control agent output and execute remote code—e.g., sampling callbacks call external LLM APIs (https://api.openai.com/v1/chat/completions and https://api.anthropic.com/v1/messages) which return assistant content used as prompts, and multiple examples invoke npx -y to fetch-and-run npm packages (e.g., @modelcontextprotocol/server-filesystem) at runtime.