mcp-builder
Pass
Audited by Gen Agent Trust Hub on Aug 18, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill specifies the official 'anthropic' and 'mcp' Python packages in 'scripts/requirements.txt'. These are well-known and trusted dependencies required for building and testing MCP-compliant servers.\n- [COMMAND_EXECUTION]: The evaluation harness in 'scripts/evaluation.py' and connection handlers in 'scripts/connections.py' utilize 'stdio' transport to launch and communicate with local MCP server processes. This subprocess execution is the intended mechanism for local server testing and is controlled by user-specified CLI arguments.\n- [INDIRECT_PROMPT_INJECTION]: The 'scripts/evaluation.py' script contains an inherent indirect prompt injection surface as it reads test questions from an external XML file and passes them to the LLM for processing. This is a functional requirement for an evaluation tool. The mandatory evidence chain is as follows:\n
- Ingestion points: Data is read from XML qa_pairs in 'scripts/evaluation.py'.\n
- Boundary markers: The 'EVALUATION_PROMPT' uses specific tags for outputs (, , ) but does not use delimiters to isolate the input questions from the instructions.\n
- Capability inventory: The script facilitates tool calls on the server under test and communicates with the Anthropic API.\n
- Sanitization: Input content is not filtered or escaped, which is standard for evaluation frameworks targeting developer-authored test cases.
Audit Metadata