mcp-builder

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

• Third-party content exposure detected (high risk: 0.90). The SKILL.md and referenced docs explicitly instruct fetching and reading public web pages (e.g., Phase 1.2: "https://modelcontextprotocol.io/sitemap.xml" and raw.githubusercontent.com SDK READMEs) and the evaluation guide (Step 1 and Step 4) requires fetching additional documentation and read-only content from the web, so the agent is expected to ingest untrusted, public third-party content that can influence tool design and actions.

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

• Potentially malicious external URL detected (high risk: 0.90). The evaluation harness and connection code explicitly allow connecting to a remote MCP server URL (e.g., https://example.com/mcp) at runtime and then calls tools via connection.call_tool, which executes code on that remote server and directly affects the agent's tool-driven behavior.