mcp-builder
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The evaluation harness in
scripts/evaluation.pyandscripts/connections.pyuses themcplibrary to launch local MCP servers as subprocesses via shell commands. This is used for developer testing and debugging. - [EXTERNAL_DOWNLOADS]: The skill fetches documentation and SDK details from the
modelcontextprotocol.iowebsite and GitHub repositories under themodelcontextprotocolorganization. - [INDIRECT_PROMPT_INJECTION]: The
scripts/evaluation.pyscript incorporates external data into the LLM prompt. 1. Ingestion points: XML evaluation files and tool outputs. 2. Boundary markers: The prompt uses XML-style tags to delimit agent responses. 3. Capability inventory: The script can execute subprocesses to interact with servers. 4. Sanitization: No explicit filtering is applied to ingested content beyond standard parsing.
Audit Metadata