Model Context Protocol (MCP) servers provide a new approach to unify automation and observability across hybrid Cisco environments. They enable an AI client to automatically discover and use tools across multiple Catalyst Center clusters and Meraki organizations.
If you’re curious about how this works, now’s the time to see it in action.
In this new demo, Cisco Principal Technical Marketing Engineer Gabi Zapodeanu shows how a single AI client routes natural-language queries to the right tool, retrieves responses from multiple domains, and helps you troubleshoot or report on your network more efficiently.
See MCP in Action: Catalyst Center and Meraki Integration
In the video below, Gabi demonstrates how MCP servers enable an AI client to interact with tools across multiple platforms. You will learn:
- How the client connects to multiple MCP servers and discovers available tools.
- How those tools are selected and executed in real time based on user intent.
- How a single query can span clusters and organizations using patterns like cluster = all.
The video includes practical walkthroughs of multi-cluster inventory lookups, issue correlation across, and a BGP troubleshooting workflow built from basic tools.
Understanding MCP Architecture and Workflow
MCP uses a client-server protocol that enables an AI assistant to connect to multiple MCP servers and dynamically discover available tool definitions. Here is what the full workflow looks like:
- An AI client, powered by a large language model, connects to multiple MCP servers.
- Each server provides a list of tools—either prebuilt runbooks or auto-generated APIs.
- A user asks a question; the AI client selects the appropriate tool, fills in the parameters, and sends the request.
- The tools execute, return data, and the AI responds to the user.
This enables asking a single question—such as “Where is this client connected?”—and receiving answers from multiple clusters and organizations.
Imperative Tools vs. Declarative Tools in MCP Servers
The demo explains two types of tools supported by MCP servers:
- Imperative tools are predefined sequences written in Ansible, Terraform, or Python. They are best suited for write tasks where guardrails and strict execution order are important.
- Declarative tools are auto-generated from YAML files and are ideal for read-heavy tasks such as inventory, event lookup, or compliance checks. They also support pagination with offset and limit parameters.
Gabi shares examples of both types, demonstrating their use in real scenarios like firmware checks and cross-domain client discovery.
Troubleshooting and Compliance Using Generative AI Flows
Beyond single-tool calls, MCP supports multi-step workflows. These generative AI flows enable you to:
- Correlate events
- Identify root causes of issues such as BGP flaps
- Run compliance checks or collect telemetry across sites
- Apply guardrails for changes, ensuring only trusted runbooks are used for configuration actions
The MCP client learns from tool usage patterns and can suggest new tools based on frequent API calls.
How to Get Started and What’s Next
This demo provides a clear, practical introduction to MCP for anyone working in NetOps or DevOps. You will gain a better understanding of:
- Why MCP matters today
- How to connect MCP to your Cisco platforms
- The types of tools and workflows it supports
- How to structure your own tools using YAML or SDKs
Watch the full replay:
Subscribe to Cisco DevNet on YouTube to get notified when new demos go live.
