mcp-builder
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches documentation and SDK information from official Model Context Protocol resources, including modelcontextprotocol.io and the modelcontextprotocol GitHub organization.
- [COMMAND_EXECUTION]: Provides scripts for evaluating local MCP servers. The transport mechanism in scripts/connections.py executes a command and arguments specified by the user to launch the server as a subprocess for testing purposes.
- [INDIRECT_PROMPT_INJECTION]: The evaluation harness in scripts/evaluation.py processes external data that could potentially contain instructions.
- Ingestion points: The parse_evaluation_file function reads task questions from a user-provided XML file.
- Boundary markers: The EVALUATION_PROMPT utilizes specific XML tags (, , ) to structure agent output and separate it from task processing.
- Capability inventory: The agent has access to tools provided by the MCP server being tested via call_tool in scripts/connections.py, and the script can perform network calls to the Anthropic API or execute local commands based on user transport configuration.
- Sanitization: The script uses specific output formatting requirements in the system prompt to guide agent behavior and filter tool responses.
Audit Metadata