Itential MCP secures AI interactions and infrastructure

The Itential Model Context Protocol (MCP) Server integrated with the Itential Platform establishes a security architecture that safeguards infrastructure operations by preventing direct access from Artificial Intelligence (AI) systems, thus addressing concerns related to uncontrolled access.

The Security Challenge

With the adoption of AI in automation, organizations face the essential question of how to allow AI assistants to effectively manage complex systems while safeguarding security. The solution is to implement a secure mediation layer that sits between AI systems and the production infrastructure.

Architecture Overview

The Itential MCP adheres to the open MCP specification, which creates a structured communication framework between Large Language Models (LLMs) and the Itential Platform. This system ensures that communications are controlled and auditable rather than allowing direct Application Programming Interface (API) access.

• Protocol Implementation: The MCP specification is implemented with stdio and Hypertext Transfer Protocol (HTTP) for standardized tool interactions.
• Authentication & Authorization: Uses JWT-based authentication with optional Open Authorization 2.0 (OAuth 2.0) for secure client-server interactions.
• Client Mediation Layer: Features a custom wrapper that standardizes HTTP methods, enhances service discovery, and ensures robust exception handling.

How Mediation Works

AI requests originate and undergo a structured security process that comprises several key stages before reaching the infrastructure.

1. AI Request Reception: Natural language requests from AI are converted into structured calls by MCP clients.
2. Authentication & Validation: The server checks JWT tokens and validates requests against tool schemas.
3. Translation to Platform Operations: AI requests are mapped to specific API endpoints on the Itential Platform.
4. Controlled Execution: The platform processes API calls with the necessary authentication under user permissions.
5. Response & Logging: All transactions are logged, and structured responses are relayed back to the AI assistant.

The AI assistant does not directly access the Itential Platform API, maintaining security through the mediation layer at all times.

Security Boundaries

The MCP enforces distinct separation between AI systems and infrastructure components across four layers.

• AI Layer: Deals with natural language inputs.
• MCP: Manages validated tool calls and authentication processes.
• Itential Platform: Handles authorized API operations and orchestration of workflows.
• Infrastructure: Involves network devices and cloud resources.

Multiple authentication boundaries provide robust access control across the process, ensuring that each layer retains its own authentication context.

Tool-Level Access Control

Granular access control is enforced by the MCP through a tagging system that allows role-based access configurations. This system employs various tags to limit scope based on user roles.

Translation Layer

The MCP facilitates the translation of unstructured AI requests into structured platform operations, ensuring appropriate actions are taken based on validated data and API requirements.

The service layer integrates functionalities that manage external service execution and validates inputs against defined schemas.

Logging & Traceability

Multi-level logging is applied to maintain comprehensive traceability, capturing all aspects of the interaction from requests to responses, thus supporting auditing and compliance evaluations.

Authentication attempts and performance metrics are monitored, contributing to the overall security framework.

Real-World Use Cases

Practical applications include AI-driven platform health monitoring, device configuration validation, and the orchestration of complex automated workflows—all ensuring secure processes without direct access to infrastructure.

Best Practices

Organizations are advised to implement access based on necessity, utilize OAuth 2.0 for production environments, and maintain secure logging and monitoring practices.

Conclusion

The Itential MCP offers a structured approach to integrating AI in network automation while ensuring strict control measures are in place to protect infrastructure access. It emphasizes secure, auditable processes with multiple layers of authentication and detailed logging for operations.

This summary reflects the core principles of operational security while enabling AI capabilities within organizational frameworks.