An MCP server that executes command-line tools in a secure and configurable manner, allowing AI-powered applications to execute commands on behalf of users
Scoring was not performed
No explicit tool definitions with proper MCP schemas, tools are inferred from dynamic YAML configuration. Tool registration is implicit and not discoverable from source code.
No documented output schemas. The CreateMCPTool function builds input parameter definitions but there is no specification of what each tool returns, the structure of results, or fields the LLM can expect.
Tool naming is completely unconstrained, YAML configuration allows arbitrary tool names. No enforcement of verb-noun conventions. Names may be 'run_command' (generic) rather than specific action verbs like 'get_', 'create_', 'update_', 'delete_', 'search_', 'list_'.
Parameters are loaded from YAML config map[string]ParamConfig, types are dynamically inferred (default to 'string' if not specified). No validation that required parameters are actually marked as required. Type inference from YAML is fragile and error-prone.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-22 | F | 33 | <=2025-11-25 | v2 |
| 2026-03-09 | F | 18 | - | v1 |
No error handling or recovery guidance visible. If a tool command fails, the response is unstructured output from the shell runner. No actionable error messages, no classification of retryable vs user-fixable errors.
Marked as DESTRUCTIVE but no confirmation step, dry-run, or confirmation-request pattern visible. Arbitrary shell commands can delete files, modify systems, or cause irreversible damage without LLM safeguards.
Tool definitions are not discoverable from source code alone, they exist only in user-provided YAML. The evaluator cannot assess tool naming, parameter descriptions, or schema completeness without example configuration files.