This is not an MCP server, it is a collection of integration hooks and scripts for third-party IDEs (Cline, Windsurf, Cursor) and CI/CD systems (GitHub Actions, Jenkins). The 6 'tools' are not MCP tool definitions but rather hook entry points and utility functions. None have formal JSON Schema input definitions visible in the source. Descriptions are present but generic. No output schemas documented. The repository provides policy-scanning integrations for Zscaler AI Guard, not a remotely-callable MCP server. Scoring reflects that tool definitions are inferred from hook signatures rather than explicit MCP registration.

• If converting to a true MCP server: register tools via the MCP protocol with explicit JSON Schema input and output definitions. Use the MCP SDK (Python mcp library) to expose tools over HTTP or STDIO.
• Rename hook entry points to action verbs: 'scan_tool_request' (PreToolUse), 'audit_tool_response' (PostToolUse), 'validate_prompt' (scan_policy). This clarifies intent for LLM tool selection.
• Add formal output schemas documenting the structure of scan results (e.g., {"blocked": bool, "reason": string, "policy_id": string}).
• Expand descriptions to 50 - 200 characters, explaining WHEN to call each tool and WHAT to do if a policy violation is detected (e.g., 'Blocks execution if content violates AI Guard policy; returns reason and policy ID for logging.').
• Document parameter constraints: direction enum values (IN, OUT), expected content types, and maximum content length. Add validation error messages that guide recovery.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry