mcp-builder

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

• Third-party content exposure detected (high risk: 0.75). The evaluation harness connects to an MCP server over SSE/HTTP (outsider-controlled URL/headers) and then ingests the server’s tool outputs (free-text result.content from connection.call_tool(...)) into the agent’s messages context via tool_result blocks, enabling indirect prompt injection.