mcp-builder
Pass
Audited by Gen Agent Trust Hub on Aug 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The evaluation script (scripts/evaluation.py) and connection utility (scripts/connections.py) facilitate the execution of local MCP servers via the stdio transport mechanism. This involves spawning a subprocess based on user-supplied commands and arguments, which is an expected capability for a tool designed to test local development servers.
- [INDIRECT_PROMPT_INJECTION]: The evaluation harness in scripts/evaluation.py implements an agent loop that ingests outputs from MCP tools and incorporates them into the LLM context. This represents a potential surface for indirect prompt injection if a tested tool provides malicious output, although this is a standard architecture for tool-using agents.
- [EXTERNAL_DOWNLOADS]: The skill instructions and documentation recommend fetching official SDK documentation and protocol specifications from modelcontextprotocol.io and the modelcontextprotocol GitHub organization. These references target official project resources and are treated as safe references to well-known services.
Audit Metadata