tavily-remote-mcp
Pass
Audited by Gen Agent Trust Hub on Mar 10, 2026
Risk Level: SAFE
Full Analysis
- [CREDENTIALS_UNSAFE]: The scripts include a hardcoded fallback API key 'BETA_TEST_KEY_MARCH_2026' for the OneKey Agent Router. This is presented as a demo key for testing purposes and is intended to be overridden by the DEEPNLP_ONEKEY_ROUTER_ACCESS environment variable.
- [DATA_EXFILTRATION]: The skill transmits user-provided search queries, URLs, and extraction parameters to a remote endpoint (deepnlp.org) via the OneKeyAgentRouter. This is the primary function of the skill and uses the vendor's infrastructure.
- [PROMPT_INJECTION]: Several tools, such as tavily_crawl and tavily_research, accept natural language instructions from the user. This represents a surface for indirect prompt injection. Ingestion points: User-provided payload in all script files. Boundary markers: Absent. Capability inventory: Network invocation via OneKeyAgentRouter.invoke in all scripts. Sanitization: Absent (delegated to the remote service).
Audit Metadata