bgpt-paper-search

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

• Third-party content exposure detected (high risk: 0.85). The required workflow uses a remote MCP server (configured via a URL) whose search_papers tool returns structured data derived from full-text scientific papers, meaning the agent will ingest outsider-authored paper content/results into its LLM context at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). The MCP configuration passes the runtime SSE endpoint https://bgpt.pro/mcp/sse to the agent (npx mcp-remote), which the agent will connect to at runtime and which can directly supply prompts/instructions and behavior-controlling messages.