# Agentic analytics over MCP: ask Claude or ChatGPT about your users

> With an MCP server, the AI app you already use becomes your analytics agent. Connect Behavly to Claude, ChatGPT, Gemini or Cursor and ask what your visitors did, who got stuck and how your campaigns went, with your role, your plan and an approval step for anything that deletes or sends.

Source: https://behavly.io/blog/agentic-analytics-mcp · Behavly blog · Agentic Analytics · November 17, 2026 · By Ardit Karavidaj, Founder & CEO, FoundryX

**MCP**, the Model Context Protocol, is the standard way for an AI app to use another product’s tools. Connect an analytics product over MCP and the AI app you already work in becomes your analytics agent: you ask in its chat, and it reads your data through the product’s own tools.

Behavly runs an MCP server. Connect it to Claude, ChatGPT, Gemini, Perplexity, Grok, Cursor, Claude Code or VS Code, and you can ask about your website users without opening Behavly: who came, where signups drop off, who got stuck, which campaigns worked. It is [agentic analytics](https://behavly.io/blog/what-is-agentic-analytics) inside the app you already use.

#### What you can ask

The AI app gets the same actions as **Behavly Agent**, Behavly’s own chat. For analytics, that means questions like these:

| You ask | The Behavly tool it uses |
| --- | --- |
| What did people do on my site this week? | Check site behavior: visitors, visits, page views, clicks, rage clicks, bounce rate, top pages, sources and a trend |
| Where do signups drop off? | Check site behavior: the signup funnel with where people dropped and which form fields failed |
| What do people click on /pricing? | Read the page’s heatmap: top elements, rage clicks, dead clicks, scroll depth |
| Who visited pricing this week? | Look up identified users: one row per person |
| How many users have gone quiet? | Check audience health: active, at risk, inactive and new |
| Find people who signed up but never came back | Find a pattern: rules in plain words and the exact count |
| How are my campaigns doing? | Campaign performance across the workspace |

Ask for a graph and the app can have Behavly draw one from the numbers it just looked up. And when an answer calls for a step, such as saving a pattern or building a campaign for the people in it, the app can do that too, inside the limits below.

**~160** tools Behavly offers an AI app over MCP: campaigns, contacts and patterns, the Unibox, templates, insights, email accounts and settings.

#### Connect it in a few minutes

1. Copy the server URL from **Settings → MCP** in Behavly.
2. In your AI app, add a new MCP server or custom connector, name it **Behavly** and paste the URL.
3. Connect. Behavly’s sign-in page opens.
4. Choose **Continue with Behavly**, pick the workspace and choose **Read & write** or **Read only**.
5. Turn Behavly on in a conversation and ask.

- **Claude**: In claude.ai or the Claude app, go to **Customize → Connectors** and click **Add custom connector**.
- **ChatGPT**: Turn on **Developer mode** under **Settings → Apps & Connectors → Advanced settings**, then click **Create** and choose OAuth. Developer mode needs a paid ChatGPT plan.
- **Claude Code**: Run the command below, then type `/mcp` and choose Behavly to sign in.

```bash
claude mcp add --transport http behavly https://behavly.io/api/mcp
```

Settings → MCP shows the steps for each app. For analytics only, choose **Read only**: lookups work and every change is refused.

#### Load only the analytics tools

Some apps load only so many tools. End the server URL with the areas you want, and the app gets just those:

```text
https://behavly.io/api/mcp?areas=insights,contacts
```

The areas are `campaigns`, `agents`, `contacts`, `inbox`, `templates`, `insights`, `mailboxes` and `settings`.

#### The limits that come with it

An AI app with access to your user data needs limits it cannot talk its way around. With Behavly they live on the server, not in the prompt:

- **Your person, your workspace, your role.** A connection acts as the person who made it, in one workspace, with that person’s role as it is right now. Change the role and the connection follows.
- **Deleting and sending wait for you.** Deleting something or sending email to real people waits for your approval in Behavly: the app gets a link, and you approve or decline within 30 minutes.
- **Your plan’s limits apply.** What the app does counts toward your plan exactly as if you did it in Behavly. MCP calls are not Behavly Agent chat messages and do not count toward them.
- **Everything is recorded.** Changes made from outside Behavly appear in the audit log as **via MCP**, with the key’s name.
- **You can cut it off at once.** Disconnect an app under **Connected apps**, revoke a key, or switch **Workspace access** off for every key. With AI switched off in **Settings → Account → AI & Privacy**, AI apps cannot connect at all.

> Note: Every Behavly plan can use MCP and the API, Free included.

#### MCP or Behavly Agent?

Both read the same data and run the same actions. Use **Behavly Agent** when you are working in Behavly: it sits on Home and beside the Inbox and Heatmaps, and changes run from a confirmation card in the chat. Use **MCP** when your day is in Claude, ChatGPT or Cursor and you want your user data next to everything else you are doing there.

For questions to start with, see [12 agentic analytics examples](https://behavly.io/blog/agentic-analytics-examples). For the full setup, see [Connect AI apps and other tools](https://behavly.io/help-center/agent/connect-ai-apps-mcp).

Start on the free plan: https://behavly.io/auth/login?mode=signup
