HPE Campus Automation for the Self-Driving Network

Learn how Model Context Protocol (MCP) enables cross-domain automation across both HPE Aruba Central and HPE Mist that simplifies IT operations. This session explores how HPE Self-Driving Network uses MCP to share context, trigger actions, and enhance network intelligence. See real-world examples of how MCP connects network intelligence with agent-driven workflows to enable more efficient, automated, and actionable network operations. Jay Pathak and Thomas Munzer of HPE’s campus and branch automation team, focused on agentic automation, presented the session at Networking Field Day 41, with Munzer joining remotely from France.

Pathak started with the automation ecosystem HPE has built around both platforms, including Ansible and Terraform for infrastructure as code, ServiceNow for automatic ticketing, and a range of third-party integrations. He said both Mist and Aruba Central are 100% API-first with standardized OpenAPI models, so anything an operator can do in the UI has an API call behind it. The platforms also offer webhooks for event-triggered workflows, WebSocket and streaming APIs for raw telemetry, Postman collections, and SDKs for Python, Go, and Ansible. Both now offer cloud-hosted MCP servers, which Pathak called the API for AI models and agents. He framed response quality as the product of MCP server quality, data quality, and LLM quality, noting that HPE controls the first two and that its servers work with lightweight or frontier models. In a conversational demo, an MCP client checked user issues in a Mist org for the week, returned recommended actions, and then traced a client to its AP and upstream switch in a single query that spanned multiple domains.

Munzer then showed a fully agentic workflow. In his setup, Aruba access points managed by Central connected to a Juniper switch managed by Mist, and both platforms sent alerts through a webhook collector to an AI agent with its own LLM, memory, and MCP servers for Aruba Central, Mist, and ServiceNow. When he disabled PoE on the switch port, Central reported the AP as disconnected, and the agent queried both platforms, correlated the data, and opened a ServiceNow incident. The incident identified the root cause as the switch configuration change in Mist, not the AP itself. Munzer argued that scripting every event combination across multiple platforms isn’t practical, while an agent can adapt to whatever it finds. The result is proactive operations, with teams learning about issues almost in real time, with full context, before users ever call support.