mcp-builder
Pass
Audited by Gen Agent Trust Hub on Jun 19, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill fetches documentation and SDK information from official Model Context Protocol repositories on GitHub. These operations are used to provide the agent with up-to-date specifications during the development process and target well-known, authoritative sources.
- [COMMAND_EXECUTION]: The skill provides a connection utility in
scripts/connections.pyand an evaluation harness inscripts/evaluation.pythat can launch local MCP servers via standard input/output (stdio). This involves executing local commands and arguments provided by the user to test the developed servers. - [INDIRECT_PROMPT_INJECTION]: The evaluation harness in
scripts/evaluation.pyprocesses data returned from external MCP servers during testing. - Ingestion points: Data enters the system through
connection.call_tool()outputs and the parsing of user-provided XML evaluation files inparse_evaluation_file. - Boundary markers: The script uses a structured system prompt (
EVALUATION_PROMPT) that requires the model to wrap its output in<summary>,<feedback>, and<response>tags to maintain context separation. - Capability inventory: The script can execute any tool exposed by the target MCP server, which may include file system or network access depending on the server's implementation.
- Sanitization: The script performs basic extraction of tagged content but does not implement complex sanitization for external tool outputs, relying on the agent's internal instruction following.
Audit Metadata