mcp-builder

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

• Third-party content exposure detected (high risk: 0.90). The guide and scripts explicitly instruct the agent/developer to fetch and load public web content (e.g., "https://modelcontextprotocol.io/sitemap.xml", raw GitHub URLs like "https://raw.githubusercontent.com/…/README.md", and "use web search and WebFetch"), and the evaluation/connection code accepts arbitrary remote MCP URLs (-u) which the agent is expected to read and interpret as part of its workflow, exposing it to untrusted third‑party content.

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

• Potentially malicious external URL detected (high risk: 0.90). The evaluation harness connects at runtime to a user-specified MCP server URL (e.g., https://example.com/mcp) and loads the server's tool definitions and results into the agent loop, meaning remote content from that URL can directly control the agent's tool-driven prompts and behavior.