mcp-builder

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

• Third-party content exposure detected (high risk: 0.90). The skill's SKILL.md workflow explicitly instructs the agent/developer to fetch and ingest external public pages (e.g., "Start with the sitemap at https://modelcontextprotocol.io/sitemap.xml" and "Use WebFetch to load https://raw.githubusercontent.com/.../README.md"), and the evaluation harness (scripts/evaluation.py and connections.py) also supports calling remote MCP servers over SSE/HTTP, so the agent is expected to read untrusted, public third‑party content that can influence tool calls and decisions.

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 an external MCP server URL (e.g., https://example.com/mcp) and calls it to load tool definitions (via list_tools) which are injected into the model as "tools" and thus directly control the agent's available prompts/actions and remote execution flow.