mcp-setup

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the claude mcp add command-line interface to register new tools, which involves executing shell commands and modifying local configuration files.
  • [EXTERNAL_DOWNLOADS]: Fetches and executes code from well-known package registries via npx, including @upstash/context7-mcp from Upstash and exa-mcp-server from Exa.
  • [EXTERNAL_DOWNLOADS]: Downloads and runs the GitHub MCP server container from the official GitHub Container Registry (ghcr.io/github/github-mcp-server).
  • [EXTERNAL_DOWNLOADS]: References an official Model Context Protocol filesystem server package (@modelcontextprotocol/server-filesystem) for extended file access.
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in the 'Custom MCP Server' configuration section where user-supplied strings are used to construct shell commands.
  • Ingestion points: User input gathered via AskUserQuestion for server names, commands, arguments, and environment variables.
  • Boundary markers: No explicit delimiters or instructions are used to prevent the agent from obeying commands embedded in user input.
  • Capability inventory: Shell command execution via claude mcp add in SKILL.md.
  • Sanitization: No input validation or escaping mechanisms are defined to sanitize user-provided strings before interpolation into shell commands.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 07:44 PM
Security Audit — agent-trust-hub — mcp-setup