AI Agent

The OSPF & IS-IS AI Agent is a natural-language assistant for your IGP. It talks to real OSPF/IS-IS domains — pulling from Topolograph snapshots or live Watcher state through the MCP server — and answers questions in plain English through a Streamlit web UI.

vadims06/ospf-isis-ai-agent Demo video

What you can ask

• Which graphs are currently connected?
• What nodes are in the latest OSPF area 0 graph?
• Which networks are assigned to a specific host?
• What is the route between two IP addresses?
• What happened to the links after the topology changed?

How it fits together

flowchart LR
    U[You<br/>natural language] --> S[Streamlit UI]
    S --> LLM[OpenAI model]
    LLM --> MCP[Topolograph MCP server]
    MCP --> T[Topolograph<br/>snapshots + live state]

The agent uses the MCP server as its bridge to Topolograph, so it can answer with grounded, real network data rather than guesses.

Quick start (Docker)

Prerequisites

An OpenAI API key (the demo costs well under \$1). A local LLM option via vLLM is in development.

Because the current setup uses public OpenAI models, a local MCP endpoint must be reachable from OpenAI — so you expose it with a tunnel.

Cloudflare Tunnelngrok

docker-compose --profile cloudflare up --build

Watch the logs for a public URL like https://your-tunnel-url.trycloudflare.com.

ngrok http 8080   # Topolograph's MCP server (published via Nginx on 8080)

Then point the agent at the tunnel in .env:

MCP_SERVER_URL=https://your-tunnel-url.trycloudflare.com

Start it and open the UI:

docker-compose up --build
# http://localhost:8501

Reproduce the demo

# 1. Topolograph + an OSPF lab
git clone https://github.com/Vadims06/topolograph-docker
cd topolograph-docker
sudo ./install.sh

# 2. The assistant (from this repo)
docker-compose up --build

Local development

cp .env.template .env   # then edit
pip install -r requirements.txt
streamlit run app.py

---

Related: MCP Server · Python SDK · Real-time monitoring