tanstack-ai-mcp

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

• Third-party content exposure detected (medium risk: 0.30). At runtime, chat({ mcp: { clients: [...] } }) performs MCP tool discovery by calling client.tools() on each configured external MCP server and forwards the server-provided JSON Schema (and related tool metadata) into the LLM’s chat() messages/tool list without any requirement to first select a specific outsider-authored item.

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

• Potentially malicious external URL detected (high risk: 1.00). The skill connects at runtime to MCP servers (e.g. https://mcp.example.com/mcp and https://mcp.github.com/mcp) and fetches prompts/resources (client.getPrompt, client.readResource) which are injected into model input, so these external URLs directly control agent prompts during execution.