mcp-builder
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/connections.pyandscripts/evaluation.pyfiles provide the ability to launch local MCP servers using thestdiotransport. This is a core functionality of the toolkit designed to allow developers to test their implementations locally. Commands and arguments are provided by the user via the command-line interface, ensuring transparency and user control. - [INDIRECT_PROMPT_INJECTION]: The
scripts/evaluation.pyscript possesses an attack surface for indirect prompt injection as it processes tool outputs from external servers and feeds them into the agent's context. However, this is inherent to the purpose of an MCP client and the Model Context Protocol itself. - Ingestion points: Tool outputs are ingested and added to the message history in
scripts/evaluation.py(within theagent_loopfunction). - Boundary markers: None specific to ignoring instructions; uses standard message roles.
- Capability inventory: The script can execute local commands via the
stdiotransport inscripts/connections.py. - Sanitization: None; outputs are serialized to JSON and passed directly to the model.
- [EXTERNAL_DOWNLOADS]: The documentation and scripts reference official Model Context Protocol resources (e.g.,
modelcontextprotocol.io,github.com/modelcontextprotocol) and the Anthropic SDK. These are well-known technology organizations and official project repositories. All external references are handled neutrally and follow standard development patterns.
Audit Metadata