⚛️ By 2026 I Talk To My Network, why & how ?

Chloé Durand
4 minutes ago
2 min read

Move from manual diagnosis to AI-Assisted Analysis

In this new edition and last 2025 edition, we will discuss how new conversational technologies can significantly improve network management costs.

Conversational network intelligence to predict, correlate, and improve productivity

Modern networks generate massive volumes of performance data, which is often scattered and complex to interpret. Every minute spent manually correlating these measurements represents a direct operational cost and a risk of delay in resolving incidents. In a context where the reliability and profitability of telecom infrastructures are strategic, the challenge now is to transform raw data into immediate and measurable decisions.

MCP, what’s this and how does it work ?

The Model Context Protocol (MCP) acts as an intelligent translator between the technical language of the network and analysis tools. It links the model, context, and protocol to give meaning to latency, throughput, and radio quality measurements.

This protocol creates a layer of intelligence that allows you to dynamically query and understand network behavior, either in natural language or via API.

Let see the top 4 reasons to implement AI Connect agent with Analyzer

Better customer reporting → lower churn
Upsell / cross-sell based on real usage data
Lower NOC / engineering workload per ticket
Reduce mean time to resolve (MTTR) by up to 70%

Okay, how does it work with Claude Code as an example ?

Step 1 – You open the chat You type in the chat box: “Analyze latency on the LEO-Backhaul-Paris link between 10 a.m. and 11 a.m. this morning, compare it with the 5G-Private-Paris link, and give me a summary of the possible causes.”

The Claude Code client calls your MCP server:

get_latency_timeseries("LEO-Backhaul-Paris", 10:00, 11:00)
get_jitter_timeseries("LEO-Backhaul-Paris", 10:00, 11:00)
same for 5G-Private-Paris
compute_stability_index(...)

Step 2 – Claude uses the MCP server connected to Latencetech-Analyzer In the conversation, you see something like: “I’m going to retrieve the latency and jitter time series for both links over this time window, then compute the stability index.”

Step 3 – Displaying the results Claude replies with an already digested answer:

“Between 10:00 and 11:00:

Link LEO-Backhaul-Paris:
- average latency: 85 ms, peaks up to 220 ms, high jitter (20–30 ms).
- stability drops to 0.62 between 10:32 and 10:41 (likely congestion / scheduling).
Link 5G-Private-Paris:
- average latency: 18 ms, very few peaks, jitter < 5 ms, stability 0.91.

Hypothesis: overload on the LEO backhaul (window 10:32–10:41), while 5G remains stable → it should be preferred for real-time traffic in this time slot.”