mcp-builder
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches official MCP specification and SDK documentation from
modelcontextprotocol.ioandgithub.com/modelcontextprotocol. These resources provide the agent with authoritative protocol information for server development. - [COMMAND_EXECUTION]: The evaluation harness (
scripts/evaluation.py) executes user-specified commands to launch local MCP servers using thestdiotransport. This functionality is triggered by command-line arguments and is essential for testing server logic in a development environment. - [INDIRECT_PROMPT_INJECTION]: The evaluation harness processes external evaluation questions from XML files and integrates outputs from the MCP server tools into the agent's context.
- Ingestion points: Question data is loaded from XML files in
scripts/evaluation.py; tool responses are retrieved from the MCP server via transport layers. - Boundary markers: The agent is guided by a specific system prompt, though the ingested text itself is not wrapped in special delimiters.
- Capability inventory: The skill can execute local processes and perform network API calls to model providers.
- Sanitization: No automated filtering or sanitization is performed on the ingested content, relying instead on the developer's oversight of test cases.
Audit Metadata