pubmed
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses the
mcp-skillandfastmcpPython packages. It connects to an official MCP server endpoint athttps://pubmed.mcp.claude.com/mcpto perform article searches and metadata retrieval. - [INDIRECT_PROMPT_INJECTION]: The skill tools process research data from the PubMed database. The implementation includes guidance for the AI agent on how to handle this data safely.
- Ingestion points: Data is ingested via MCP tool calls in all methods of the
PubmedAppclass inapp.py. - Boundary markers: Docstrings in
app.pycontain explicit instructions for the model to cite PubMed and DOIs, and to restrict tool usage to biomedical topics. - Capability inventory: The skill performs network operations using the
fastmcpclient to interact with the PubMed MCP server. - Sanitization: Tool responses are processed using
json.loadsto ensure structured data handling.
Audit Metadata