MCP Integration

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

• Third-party content exposure detected (medium risk: 0.65). The required runtime workflow is to start/connect MCP servers (stdio/SSE/HTTP/WS) and register/discover their tools, and the servers’ tool schemas/descriptions and responses are then provided to the agent/LLM context—so if those tools originate from an outside/hosted server, their content can include attacker-controlled free text (indirect prompt injection risk).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill includes runtime MCP server endpoints (e.g., https://mcp.asana.com/sse) which the agent will connect to at runtime to discover/register tools and fetch tool schemas/responses that directly control agent behavior and can execute remote tool logic, so these URLs are runtime dependencies that influence prompts/instruction execution.