mcp-builder
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes Python scripts (connections.py and evaluation.py) designed to launch and communicate with local MCP servers. These scripts use the stdio transport to execute user-specified commands. This is a documented core feature for local integration testing.\n- [INDIRECT_PROMPT_INJECTION]: The evaluation framework (evaluation.py) ingests data from external XML files containing questions that are subsequently processed by an LLM.\n
- Ingestion points: The parse_evaluation_file function in scripts/evaluation.py reads content from a user-specified XML file path.\n
- Boundary markers: The system prompt (EVALUATION_PROMPT) uses summary, feedback, and response tags to structure and delimit the AI's output.\n
- Capability inventory: The evaluator agent is granted access to the tools exposed by the MCP server being tested.\n
- Sanitization: The script performs XML parsing but does not include explicit sanitization of the input text before passing it to the LLM context, which is expected behavior for a test harness.
Audit Metadata