---
title: "Ask Claude About Your Phone Calls"
date: "2026-10-09"
author: "CX-Engine Team"
excerpt: "With the Insights AI MCP server, Claude, Claude Code or Cursor can answer questions about your calls: volumes, answer rates, reasons for calling, with the call ids to back every answer."
category: "AI"
order: 8
alternate_slug: interroger-claude-sur-vos-appels
---

## Your Team Already Asks an AI Assistant Everything Else

If your operations or management team uses Claude, Cursor or another AI assistant, they already ask it to summarize documents, draft emails and query spreadsheets. Phone calls are usually missing from that picture. The data sits in the PBX and in your analytics tools, and getting an answer means opening a dashboard, exporting a file or asking someone who knows where to look.

Insights AI now includes an **MCP server**. MCP (*Model Context Protocol*) is the open standard that lets AI assistants use external tools. Once connected, your assistant can query your calls directly, in the same conversation as everything else.

## What the Assistant Can Reach

The server exposes 17 tools, all read-only. You never call them by name: the assistant picks them based on your question.

**On the calls analyzed by Insights AI**, it can count and compare calls by subject, agent, queue, dialed number, outcome or sentiment, search them, pull samples of reasons for calling and frustration reasons, produce an agent report or a quality audit report, find topics close in meaning to a phrase, and read the full transcript of a call.

**With [Expert Statistics](/en/products/expert-statistics)**, nine more tools cover every call that goes through the phone system, analyzed or not: queue reports (answer rate, abandoned calls, waiting times), trends over time, agent presence, the call log with each call's route through the PBX (queues, agents, IVR, voicemail), top and repeat callers, calls per agent, a report per dialed number, outbound activity and the live status of each agent.

The server tells the assistant to use the statistics tools for "how many" and "how long", the analysis tools for "why", to say which source backs each figure, and to cite the ids of the calls behind its statements.

## Prompts That Work

Some examples, and what happens behind them.

**"How many calls did we miss on the support queue last week, and what were the callers who rang back calling about?"**
The assistant fetches the queue report for the week, lists the repeat callers, then reads the analysis of their calls. You get the figures, the recurring reasons, and the call ids to check them.

**"At what time of day do we lose the most calls, and were enough agents available?"**
It crosses the hourly trend of answered calls with the agents' presence over the same period. The answer shows whether the problem is volume or staffing.

**"Which subjects caused the most frustration this month? Give me three examples for the worst one."**
It counts calls by subject with their frustration rate, then samples the frustration reasons and quotes a few calls with their ids.

**"What happened to the call from +33 6 12 34 56 78 yesterday around 10am?"**
It finds the call in the log, shows its route through the PBX (which queue, who it rang, who answered, how long it waited) and, if the call was analyzed, opens its transcript.

**"Prepare a coaching sheet for Julie: her weakest audit criteria, with call examples."**
It runs the agent report and the audit report, then picks representative calls.

The usual assistant features still apply. In Claude Code or Cursor, you can ask it to turn the answer into a CSV, a chart script or a weekly summary file.

## The Security Model

Giving an AI assistant access to call data deserves a close look at what it can and cannot do.

- **Read-only**: no tool can modify anything in CX-Engine
- **One token per PBX**: each access token is tied to a single phone system. A company with several PBXs creates one token per PBX
- **Tied to a person**: a token stops working if its owner loses access to the account or the permission to use Insights AI. The Expert Statistics tools also require the Expert Statistics permission
- **Under your control**: a token can be disabled (and re-enabled) or revoked at any time, and the **Last used** column shows when each one was last used
- **Shown once**: the token is displayed at creation and never again. If it is lost, you create a new one and revoke the old one

One point to check with your company policy: when you use your own assistant, the information needed to answer (statistics, summaries, transcript excerpts) is sent to that assistant and processed under its publisher's terms.

## Connecting Takes a Few Minutes

1. In CX-Engine, open **CRM Connector** > **Insights AI (MCP)** and create an access token for the PBX you want to query
2. In Claude, add a custom connector with the server address shown in CX-Engine, select **No sign-in**, and send the token in the `x-api-key` request header
3. Enable the connector in a conversation and ask your question

For Claude Code, it is a single command:

```bash
claude mcp add --transport http insights-ai https://your-workspace.cx-engine.com/mcp/insights --header "X-API-Key: your-token"
```

Cursor, Claude Desktop and other MCP clients use a short JSON configuration. The full steps are in [Connect your assistant (MCP)](/en/docs/insights-ai/connect-your-ai).

ChatGPT is not supported yet: it only accepts connectors that sign in with OAuth, which is coming soon.

## Not Using Claude? The Assistant Is Built In

Not everyone on the team works in an AI assistant, and that's fine. The same tools power the assistant built into [Insights AI](/en/products/insights-ai): a manager asks the same questions inside CX-Engine, gets answers that cite the calls involved, and can export them to Word. The MCP server is simply another door to the same data, for the teams who already live in their AI tools.

[Request a demo →](/en/contact?audience=end_user)
