Hey there, everybody, and welcome to the most recent installment of “Hank shares his AI journey.” 🙂 Synthetic Intelligence (AI) continues to be all the trend, and getting back from Cisco Dwell in San Diego, I used to be excited to dive into the world of agentic AI.
With bulletins like Cisco’s personal agentic AI resolution, AI Canvas, in addition to discussions with companions and different engineers about this subsequent section of AI potentialities, my curiosity was piqued: What does this all imply for us community engineers? Furthermore, how can we begin to experiment and find out about agentic AI?
I started my exploration of the subject of agentic AI, studying and watching a variety of content material to realize a deeper understanding of the topic. I gained’t delve into an in depth definition on this weblog, however listed below are the fundamentals of how I give it some thought:
Agentic AI is a imaginative and prescient for a world the place AI doesn’t simply reply questions we ask, but it surely begins to work extra independently. Pushed by the targets we set, and using entry to instruments and techniques we offer, an agentic AI resolution can monitor the present state of the community and take actions to make sure our community operates precisely as supposed.
Sounds fairly darn futuristic, proper? Let’s dive into the technical features of the way it works—roll up your sleeves, get into the lab, and let’s study some new issues.
What are AI “instruments?”
The very first thing I wished to discover and higher perceive was the idea of “instruments” inside this agentic framework. As you could recall, the LLM (giant language mannequin) that powers AI techniques is actually an algorithm skilled on huge quantities of information. An LLM can “perceive” your questions and directions. On its personal, nevertheless, the LLM is proscribed to the information it was skilled on. It could possibly’t even search the online for present film showtimes with out some “instrument” permitting it to carry out an internet search.
From the very early days of the GenAI buzz, builders have been constructing and including “instruments” into AI purposes. Initially, the creation of those instruments was advert hoc and diverse relying on the developer, LLM, programming language, and the instrument’s purpose. However lately, a brand new framework for constructing AI instruments has gotten a whole lot of pleasure and is beginning to turn out to be a brand new “normal” for instrument improvement.
This framework is called the Mannequin Context Protocol (MCP). Initially developed by Anthropic, the corporate behind Claude, any developer to make use of MCP to construct instruments, known as “MCP Servers,” and any AI platform can act as an “MCP Consumer” to make use of these instruments. It’s important to keep in mind that we’re nonetheless within the very early days of AI and AgenticAI; nevertheless, at the moment, MCP seems to be the method for instrument constructing. So I figured I’d dig in and determine how MCP works by constructing my very own very fundamental NetAI Agent.
I’m removed from the primary networking engineer to wish to dive into this house, so I began by studying a few very useful weblog posts by my buddy Kareem Iskander, Head of Technical Advocacy in Study with Cisco.
These gave me a jumpstart on the important thing matters, and Kareem was useful sufficient to supply some instance code for creating an MCP server. I used to be able to discover extra by myself.
Creating an area NetAI playground lab
There isn’t a scarcity of AI instruments and platforms at the moment. There may be ChatGPT, Claude, Mistral, Gemini, and so many extra. Certainly, I make the most of a lot of them usually for varied AI duties. Nevertheless, for experimenting with agentic AI and AI instruments, I wished one thing that was 100% native and didn’t depend on a cloud-connected service.
A main motive for this want was that I wished to make sure all of my AI interactions remained solely on my pc and inside my community. I knew I might be experimenting in a wholly new space of improvement. I used to be additionally going to ship knowledge about “my community” to the LLM for processing. And whereas I’ll be utilizing non-production lab techniques for all of the testing, I nonetheless didn’t like the concept of leveraging cloud-based AI techniques. I might really feel freer to study and make errors if I knew the danger was low. Sure, low… Nothing is totally risk-free.
Fortunately, this wasn’t the primary time I thought of native LLM work, and I had a few potential choices able to go. The primary is Ollama, a strong open-source engine for working LLMs regionally, or no less than by yourself server. The second is LMStudio, and whereas not itself open supply, it has an open supply basis, and it’s free to make use of for each private and “at work” experimentation with AI fashions. Once I learn a current weblog by LMStudio about MCP help now being included, I made a decision to offer it a attempt for my experimentation.
LMStudio is a shopper for working LLMs, but it surely isn’t an LLM itself. It offers entry to a lot of LLMs out there for obtain and working. With so many LLM choices out there, it may be overwhelming if you get began. The important thing issues for this weblog put up and demonstration are that you just want a mannequin that has been skilled for “instrument use.” Not all fashions are. And moreover, not all “tool-using” fashions truly work with instruments. For this demonstration, I’m utilizing the google/gemma-2-9b mannequin. It’s an “open mannequin” constructed utilizing the identical analysis and tooling behind Gemini.
The subsequent factor I wanted for my experimentation was an preliminary concept for a instrument to construct. After some thought, I made a decision a great “hiya world” for my new NetAI mission could be a approach for AI to ship and course of “present instructions” from a community machine. I selected pyATS to be my NetDevOps library of selection for this mission. Along with being a library that I’m very accustomed to, it has the good thing about automated output processing into JSON by the library of parsers included in pyATS. I might additionally, inside simply a few minutes, generate a fundamental Python perform to ship a present command to a community machine and return the output as a place to begin.
Right here’s that code:
def send_show_command(
command: str,
device_name: str,
username: str,
password: str,
ip_address: str,
ssh_port: int = 22,
network_os: Non-obligatory[str] = "ios",
) -> Non-obligatory[Dict[str, Any]]:
# Construction a dictionary for the machine configuration that may be loaded by PyATS
device_dict = {
"gadgets": {
device_name: {
"os": network_os,
"credentials": {
"default": {"username": username, "password": password}
},
"connections": {
"ssh": {"protocol": "ssh", "ip": ip_address, "port": ssh_port}
},
}
}
}
testbed = load(device_dict)
machine = testbed.gadgets[device_name]
machine.join()
output = machine.parse(command)
machine.disconnect()
return output
Between Kareem’s weblog posts and the getting-started information for FastMCP 2.0, I realized it was frighteningly straightforward to transform my perform into an MCP Server/Device. I simply wanted so as to add 5 strains of code.
from fastmcp import FastMCP
mcp = FastMCP("NetAI Good day World")
@mcp.instrument()
def send_show_command()
.
.
if __name__ == "__main__":
mcp.run()
Nicely.. it was ALMOST that straightforward. I did must make a number of changes to the above fundamentals to get it to run efficiently. You possibly can see the full working copy of the code in my newly created NetAI-Studying mission on GitHub.
As for these few changes, the adjustments I made have been:
- A pleasant, detailed docstring for the perform behind the instrument. MCP purchasers use the main points from the docstring to grasp how and why to make use of the instrument.
- After some experimentation, I opted to make use of “http” transport for the MCP server fairly than the default and extra widespread “STDIO.” The rationale I went this manner was to arrange for the subsequent section of my experimentation, when my pyATS MCP server would possible run throughout the community lab surroundings itself, fairly than on my laptop computer. STDIO requires the MCP Consumer and Server to run on the identical host system.
So I fired up the MCP Server, hoping that there wouldn’t be any errors. (Okay, to be sincere, it took a few iterations in improvement to get it working with out errors… however I’m doing this weblog put up “cooking present model,” the place the boring work alongside the best way is hidden. 😉
python netai-mcp-hello-world.py ╭─ FastMCP 2.0 ──────────────────────────────────────────────────────────────╮ │ │ │ _ __ ___ ______ __ __ _____________ ____ ____ │ │ _ __ ___ / ____/___ ______/ /_/ |/ / ____/ __ |___ / __ │ │ _ __ ___ / /_ / __ `/ ___/ __/ /|_/ / / / /_/ / ___/ / / / / / │ │ _ __ ___ / __/ / /_/ (__ ) /_/ / / / /___/ ____/ / __/_/ /_/ / │ │ _ __ ___ /_/ __,_/____/__/_/ /_/____/_/ /_____(_)____/ │ │ │ │ │ │ │ │ 🖥️ Server title: FastMCP │ │ 📦 Transport: Streamable-HTTP │ │ 🔗 Server URL: http://127.0.0.1:8002/mcp/ │ │ │ │ 📚 Docs: https://gofastmcp.com │ │ 🚀 Deploy: https://fastmcp.cloud │ │ │ │ 🏎️ FastMCP model: 2.10.5 │ │ 🤝 MCP model: 1.11.0 │ │ │ ╰────────────────────────────────────────────────────────────────────────────╯ [07/18/25 14:03:53] INFO Beginning MCP server 'FastMCP' with transport 'http' on http://127.0.0.1:8002/mcp/server.py:1448 INFO: Began server course of [63417] INFO: Ready for software startup. INFO: Software startup full. INFO: Uvicorn working on http://127.0.0.1:8002 (Press CTRL+C to give up)
The subsequent step was to configure LMStudio to behave because the MCP Consumer and connect with the server to have entry to the brand new “send_show_command” instrument. Whereas not “standardized, “most MCP Purchasers use a really widespread JSON configuration to outline the servers. LMStudio is considered one of these purchasers.
Wait… if you happen to’re questioning, ‘Wright here’s the community, Hank? What machine are you sending the ‘present instructions’ to?’ No worries, my inquisitive pal: I created a quite simple Cisco Modeling Labs (CML) topology with a few IOL gadgets configured for direct SSH entry utilizing the PATty characteristic.
Let’s see it in motion!
Okay, I’m positive you might be able to see it in motion. I do know I positive was as I used to be constructing it. So let’s do it!
To start out, I instructed the LLM on how to hook up with my community gadgets within the preliminary message.
I did this as a result of the pyATS instrument wants the handle and credential data for the gadgets. Sooner or later I’d like to take a look at the MCP servers for various supply of reality choices like NetBox and Vault so it might “look them up” as wanted. However for now, we’ll begin easy.
First query: Let’s ask about software program model information.
You possibly can see the main points of the instrument name by diving into the enter/output display screen.
That is fairly cool, however what precisely is occurring right here? Let’s stroll by the steps concerned.
- The LLM shopper begins and queries the configured MCP servers to find the instruments out there.
- I ship a “immediate” to the LLM to contemplate.
- The LLM processes my prompts. It “considers” the completely different instruments out there and in the event that they may be related as a part of constructing a response to the immediate.
- The LLM determines that the “send_show_command” instrument is related to the immediate and builds a correct payload to name the instrument.
- The LLM invokes the instrument with the correct arguments from the immediate.
- The MCP server processes the known as request from the LLM and returns the outcome.
- The LLM takes the returned outcomes, together with the unique immediate/query as the brand new enter to make use of to generate the response.
- The LLM generates and returns a response to the question.
This isn’t all that completely different from what you would possibly do if you happen to have been requested the identical query.
- You’d contemplate the query, “What software program model is router01 working?”
- You’d take into consideration the other ways you might get the knowledge wanted to reply the query. Your “instruments,” so to talk.
- You’d determine on a instrument and use it to assemble the knowledge you wanted. In all probability SSH to the router and run “present model.”
- You’d assessment the returned output from the command.
- You’d then reply to whoever requested you the query with the correct reply.
Hopefully, this helps demystify just a little about how these “AI Brokers” work underneath the hood.
How about another instance? Maybe one thing a bit extra advanced than merely “present model.” Let’s see if the NetAI agent can assist establish which swap port the host is linked to by describing the fundamental course of concerned.
Right here’s the query—sorry, immediate, that I undergo the LLM:
What we should always discover about this immediate is that it’s going to require the LLM to ship and course of present instructions from two completely different community gadgets. Identical to with the primary instance, I do NOT inform the LLM which command to run. I solely ask for the knowledge I want. There isn’t a “instrument” that is aware of the IOS instructions. That data is a part of the LLM’s coaching knowledge.
Let’s see the way it does with this immediate:
And have a look at that, it was in a position to deal with the multi-step process to reply my query. The LLM even defined what instructions it was going to run, and the way it was going to make use of the output. And if you happen to scroll again as much as the CML community diagram, you’ll see that it accurately identifies interface Ethernet0/2 because the swap port to which the host was linked.
So what’s subsequent, Hank?
Hopefully, you discovered this exploration of agentic AI instrument creation and experimentation as attention-grabbing as I’ve. And possibly you’re beginning to see the probabilities on your personal each day use. For those who’d wish to attempt a few of this out by yourself, you could find every little thing you want on my netai-learning GitHub mission.
- The mcp-pyats code for the MCP Server. You’ll discover each the straightforward “hiya world” instance and a extra developed work-in-progress instrument that I’m including further options to. Be happy to make use of both.
- The CML topology I used for this weblog put up. Although any community that’s SSH reachable will work.
- The mcp-server-config.json file which you could reference for configuring LMStudio
- A “System Immediate Library” the place I’ve included the System Prompts for each a fundamental “Mr. Packets” community assistant and the agentic AI instrument. These aren’t required for experimenting with NetAI use circumstances, however System Prompts might be helpful to make sure the outcomes you’re after with LLM.
A few “gotchas” I wished to share that I encountered throughout this studying course of, which I hope would possibly prevent a while:
First, not all LLMs that declare to be “skilled for instrument use” will work with MCP servers and instruments. Or no less than those I’ve been constructing and testing. Particularly, I struggled with Llama 3.1 and Phi 4. Each appeared to point they have been “instrument customers,” however they did not name my instruments. At first, I assumed this was because of my code, however as soon as I switched to Gemma 2, they labored instantly. (I additionally examined with Qwen3 and had good outcomes.)
Second, when you add the MCP Server to LMStudio’s “mcp.json” configuration file, LMStudio initiates a connection and maintains an energetic session. Which means if you happen to cease and restart the MCP server code, the session is damaged, providing you with an error in LMStudio in your subsequent immediate submission. To repair this concern, you’ll must both shut and restart LMStudio or edit the “mcp.json” file to delete the server, reserve it, after which re-add it. (There may be a bug filed with LMStudio on this downside. Hopefully, they’ll repair it in an upcoming launch, however for now, it does make improvement a bit annoying.)
As for me, I’ll proceed exploring the idea of NetAI and the way AI brokers and instruments could make our lives as community engineers extra productive. I’ll be again right here with my subsequent weblog as soon as I’ve one thing new and attention-grabbing to share.
Within the meantime, how are you experimenting with agentic AI? Are you excited in regards to the potential? Any solutions for an LLM that works nicely with community engineering data? Let me know within the feedback under. Discuss to you all quickly!
Join Cisco U. | Be part of the Cisco Studying Community at the moment without cost.
Study with Cisco
X | Threads | Fb | LinkedIn | Instagram | YouTube
Use #CiscoU and #CiscoCert to affix the dialog.
Share: