kindly-web-search-mcp-server

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

• Third-party content exposure detected (high risk: 1.00). The skill's documented MCP tools web_search and get_content explicitly fetch and return full content from public, user-generated sources—e.g., StackOverflow threads, GitHub issues/discussions, Wikipedia, arXiv, and arbitrary webpages—so untrusted third-party text is ingested and intended to be read/used by the agent (see the "MCP Tools" and "Specialized Content Extraction" sections in SKILL.md).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly runs and installs remote code at runtime—for example the MCP startup uses "uvx --from git+https://github.com/Shelpuk-AI-Technology-Consulting/kindly-web-search-mcp-server" which fetches and executes the GitHub repo, and the installer "curl -LsSf https://astral.sh/uv/install.sh | sh" executes a remote script—both of which allow execution of externally fetched code and thus are high-risk.