mcp-builder

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

• Third-party content exposure detected (high risk: 0.70). The required runtime workflow is the evaluation harness (scripts/evaluation.py) which sends the evaluation question/expected answer to the LLM and then iteratively includes tool outputs back into the LLM context; those tool outputs come from the connected MCP server via connection.call_tool(...) (e.g., SSE/HTTP/stdio), so if the MCP server returns outsider-authored free text (e.g., scraped web content or messages from other users), it will be ingested into the agent’s LLM context through the tool_result message content.