tavily-crawl
Pass
Audited by Gen Agent Trust Hub on Mar 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/crawl.shscript executes several system utilities includingcurl,jq,base64, andsedfor API communication and data formatting. - [EXTERNAL_DOWNLOADS]: The skill downloads the
mcp-remotetool from the npm registry vianpxand makes outbound requests to official Tavily API domains (api.tavily.com,mcp.tavily.com). - [REMOTE_CODE_EXECUTION]: The script uses
npx -y mcp-remoteto execute a remote tool at runtime to facilitate the OAuth authentication flow. - [DATA_EXFILTRATION]: The script accesses the
~/.mcp-auth/directory to read and validate authentication tokens. This is part of the standard Model Context Protocol (MCP) workflow for maintaining service sessions. - [PROMPT_INJECTION]: The skill ingests untrusted data from external websites during the crawl process. This content is saved as local markdown, representing an Indirect Prompt Injection surface should the agent later read and follow instructions contained within that content.
Audit Metadata