mcp-builder
Pass
Audited by Gen Agent Trust Hub on Apr 11, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill provides legitimate development resources, including documentation for Python and TypeScript MCP implementations that follow industry best practices.
- [COMMAND_EXECUTION]: The connection utility (scripts/connections.py) enables the execution of local MCP servers as subprocesses using standard I/O. This is the intended behavior for local MCP integrations and is initiated by the user via the evaluation script's command-line interface.
- [EXTERNAL_DOWNLOADS]: The documentation suggests fetching official MCP SDK documentation and best practices from GitHub and modelcontextprotocol.io. These are well-known, trusted sources for the protocol being implemented.
- [PROMPT_INJECTION]: The evaluation harness (scripts/evaluation.py) processes outputs from MCP tools, representing a surface for indirect prompt injection. 1. Ingestion points: Tool results are ingested in the agent_loop of the evaluation script. 2. Boundary markers: The implementation uses standard SDK patterns for separating tool results. 3. Capability inventory: The script can execute subprocesses for local servers and interact with the Anthropic API. 4. Sanitization: Tool results are processed as string or JSON data without additional filtering. The risk is considered low given its intended use as a developer tool.
Audit Metadata