MCP for Cisco ACI – AI-Driven Troubleshooting

Built a minimal MCP (Model Context Protocol) architecture for cisco APIC to enable AI-driven network troubleshooting

Use Case :
 Why is endpoint in ACI Fabric  10.1.1.5 unreachable?

 How the System Thinks (End-to-End Flow) to get the root caus of the issue.

    •   AI Agent converts user query into structured MCP tool calls:

 

• MCP queries via API Call to ACI (APIC APIs) to fetch the details :
• Endpoint location (Leaf, Interface, EPG)

   Informations are recieved from the APIC

• MCP retrieves real-time faults from ACI using the API call 

• MCP queries InfluxDB for historical trends of the interface and endpoint

        Query:

• Interface status over time
• Endpoint presence history

Not a one-time issue → unstable link

• Correlation engine combines:
• Endpoint mapping
• Interface faults
• Historical behavior

 Root Cause Identified

• Endpoint mapped to Leaf-101 / Eth1/23
• Interface found DOWN with critical fault
• Historical data shows flapping behavior