mcp-builder
Pass
Audited by Gen Agent Trust Hub on Mar 6, 2026
Risk Level: SAFE
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill uses the WebFetch tool to retrieve protocol documentation and SDK information from the official Model Context Protocol website (modelcontextprotocol.io) and its associated GitHub repository (github.com/modelcontextprotocol). These are well-known, authoritative sources for the technology being integrated.
- [COMMAND_EXECUTION]: The evaluation scripts (
scripts/connections.pyandscripts/evaluation.py) are designed to run local MCP servers as subprocesses. This allows developers to test their own code using standard input/output (stdio). This execution is controlled by the user via command-line arguments and is the primary intended function of the toolkit. - [INDIRECT_PROMPT_INJECTION]: The evaluation harness in
scripts/evaluation.pyidentifies a typical attack surface for tool-using agents by processing data returned from external MCP servers. However, the skill implements boundary markers in its system prompt (XML tags like and ) to help the agent distinguish between task logic and tool output, which is a recommended best practice. - [PROMPT_INJECTION]: The instructions in
SKILL.mdprovide guidance on building tools for LLMs. No malicious instructions attempting to bypass safety filters or override system constraints were detected.
Audit Metadata