mcp-builder

Pass

Audited by Gen Agent Trust Hub on Mar 1, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches official protocol specifications and SDK documentation from modelcontextprotocol.io and the Model Context Protocol GitHub organization. These are official and trusted sources for the technology integrated by the skill.\n- [COMMAND_EXECUTION]: The evaluation harness (scripts/evaluation.py) executes local MCP server processes via the standard I/O transport. This is an intended functionality designed to allow developers to host and test their server implementations locally.\n- [PROMPT_INJECTION]: The scripts/evaluation.py script is vulnerable to indirect prompt injection through the processing of test questions from external XML files.\n
  • Ingestion points: Questions are read from the XML file provided via the eval_file positional argument in scripts/evaluation.py.\n
  • Boundary markers: None; the question content is directly interpolated into the message history sent to the model.\n
  • Capability inventory: The script is capable of spawning local subprocesses (the MCP server) and invoking any tool functionality exposed by that server.\n
  • Sanitization: None; the script does not validate or filter the content of the XML questions before they are processed by the language model.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 1, 2026, 05:31 PM