While you first hear about MCP — Mannequin Context Protocol, it feels like one thing constructed for hardcore AI researchers. However right here’s the fact: Community engineers and automation engineers are going to be a number of the largest customers of it.
In case you’re questioning why: MCP is the way you make Massive Language Fashions (LLMs) perceive your community, your topology, your requirements, your world.
With out it? You’re simply getting generic ChatGPT solutions.
With it? You’re creating Agentic AI that may configure, troubleshoot, and design networks with you.
I’ve been speaking to you — You! …Sure, you! — about community automation and adopting automation in your community engineering for years now. All in all, it’s time so as to add one other brick in *your* wall (of tech instruments). On this AI Break, we’ll discover an instance that demonstrates the worth of utilizing MCP to grasp automation in at present’s AI world.
Okay, so what’s MCP?
At its coronary heart, Mannequin Context Protocol is about injecting structured data into an LLM at runtime — robotically and programmatically.
As a substitute of manually pasting community diagrams or config templates right into a chat window, MCP lets your instruments inform the mannequin:
- What units are on the community
- What requirements you utilize
- What applied sciences you like (OSPF over EIGRP, EVPN over VXLAN, no matter)
- What change management processes exist
All that context flows into the mannequin, making its responses smarter, extra aligned, and extra helpful on your atmosphere.
Let’s begin with a fundamental, real-world instance
Let’s say you’re constructing an LLM-based Community Assistant that helps generate configs. You don’t need it suggesting RIP when your whole community runs OSPF and BGP.
With MCP, earlier than you even ask the mannequin for a config, you present AI with the next context:
Look acquainted? Yup, it’s a JSON.
{
"network_standards": {
"routing_protocols": ["OSPF", "BGP"],
"preferred_encapsulation": "VXLAN",
"security_policies": {
"ssh_required": true,
"telnet_disabled": true
}
},
"topology": {
"core_devices": ["core-sw1", "core-sw2"],
"edge_devices": ["edge-fw1", "edge-fw2"],
"site_layout": "hub and spoke"
}
}
Your assistant robotically sends this context to the LLM utilizing MCP, and then asks, “Generate a config to onboard a brand new web site.”
The mannequin now solutions in a manner that matches your atmosphere— not some random textbook response.
So, what expertise do you might want to use MCP?
Actually, lots of you have already got most of what’s wanted:
- API Fundamentals. You’ll be sending structured context (often JSON) over API calls — identical to RESTCONF, NETCONF, Catalyst Middle, Or Meraki APIs.
- Understanding your community metadata. You’ll want to know what issues: routing, VLANs, safety, machine sorts, and methods to symbolize that as structured knowledge.
- Python scripting. You’ll most likely use Python to gather this data dynamically (like through Nornir, Netmiko, or native APIs) after which bundle it into MCP calls.
- LLM fundamentals. You’ll want to perceive how prompts and context home windows work, and the way larger context equals smarter outputs.
The underside line
MCP isn’t some “perhaps later” factor for networkers.
It’s turning into the bridge between your real-world community data and AI’s potential that can assist you quicker, higher, and extra precisely.
Engineers who know methods to feed actual context into LLMs will dominate community design, troubleshooting, safety auditing, and even full-stack automation.
Begin now
- Map your community requirements.
- Package deal them as JSON.
- Play with sending that context into small AI workflows.
The perfect AI Brokers are constructed by engineers who know their community—and know methods to educate it to their AI. Subsequent, let’s get hands-on with MCP!
Attempt it
For a totally working code and directions to get began, try my undertaking on GitHub.
Create a actual Mannequin Context Protocol (MCP) server designed for community engineers.
This MCP app does the next:
- Serve your community requirements (routing protocols, safety insurance policies, and many others.)
- Reply with machine well being
- Connect with Claude Desktop, making your AI assistant conscious of your actual community atmosphere
And it’s so simple as:
- Import the MCP Python SDK
from mcp.server.fastmcp import FastMCP
- Initialize the FastMCP server with a singular identify
mcp = FastMCP("network-assistant")
- Outline instruments.
Instruments are a strong primitive within the Mannequin Context Protocol (MCP). They let your server expose actual actions—so the mannequin can question techniques, run logic, or kick off workflows. In our use case, we have to outline ‘network-standards’ & ‘machine standing’ features:@mcp.device() async def get_network_standards() -> dict[str, Any]: """Returns normal routing protocols, encapsulation, and safety insurance policies.""" return NETWORK_STANDARDS
- Run the server, and you might be set!
if __name__ == "__main__": mcp.run(transport="stdio")
And if we have a look at it, that is what the LLM is aware of about your community earlier than you contextualized it:
And that is after connecting the LLM to our Community:
The place community automation and AI really collide
You’re not scripting for the sake of scripting. And also you don’t simply use AI for the sake of buzzwords. When you may mix reside community state with LLM intelligence, you’re constructing techniques that suppose, adapt, and help with you—not simply for you.
Begin easy. Construct one stream.
Make your AI agent truly know your community. As a result of the long run belongs to engineers who don’t simply automate—they contextualize.
Welcome to the brand new frontier of Agentic AI!
Get began with AI
Studying Paths, programs, free tutorials, and extra. Unlock the way forward for know-how with synthetic intelligence coaching in Cisco U. Discover AI studying and begin constructing your expertise at present.
Join Cisco U. | Be part of the Cisco Studying Community at present without cost.
Observe Cisco Studying & Certifications
X | Threads | Fb | LinkedIn | Instagram | YouTube
Use #CiscoU and #CiscoCert to affix the dialog.
Adaptability: The Should-Have Ability for Community Engineers within the AI Period
MCP for DevOps, NetOps, and SecOps: Actual-World Use Instances and Future Insights
Share:
