Ask Claude or ChatGPT about your sales, with your real numbers

You can paste a spreadsheet into an AI chat and ask what’s going on. It sort of works. But the data is out of date the moment you export it, it’s missing the visits behind the sales, and the assistant fills the gaps with guesses that sound confident.

There’s a better way: let the assistant read your numbers itself, whenever you ask, through a connection that can’t change anything. Then “where am I losing buyers?” gets an answer built from your actual visits, funnels and sales.

How it works

Claude, ChatGPT and a growing list of AI apps can connect to outside services through MCP, the Model Context Protocol. It’s an open standard: a service describes what it can look up, and the assistant asks it when a question needs it.

odo runs an MCP server. Connect your assistant to it once, and it can read your earnings, visits, channels, pages, funnels and experiments, and only yours. It can’t change your settings, can’t move money, and never sees your API keys.

Connect it

Your MCP address is in odo under Settings, in the AI assistants group. On odo.gg it’s https://odo.gg/mcp.

• Claude (web, desktop and mobile): Settings → Connectors → Add custom connector. Paste the address, then sign in to odo and allow it.
• ChatGPT, Cursor, VS Code and others: add a remote MCP server with the same address. They open odo for you to sign in.
• Claude Code: run claude mcp add --transport http odo https://odo.gg/mcp, then /mcp to sign in.

When you allow it, you can limit the assistant to one project. Every connected app is listed in that same settings group with when it was last used, and disconnecting one cuts it off at once.

What to ask

Start with the questions you’d ask a sharp friend who could see your dashboard:

• “How’s today going compared with a normal day?”
• “What brings paying customers, and what should I stop doing?”
• “Where am I losing buyers between the product page and checkout?”
• “What’s broken or annoying on my site this week?”
• “Is my headline test done, and which one won?”
• “Review my last seven days against the week before, and give me three things to try.”

odo also hands your assistant ready-made prompts for the common jobs: a daily briefing, a weekly review, finding conversion leaks, fixing what frustrates visitors, reviewing experiments, and reviewing a single page. Apps that support MCP prompts list them for you.

What it can dig into

Behind those questions are tools odo gives the assistant. A few do most of the heavy lifting:

• find_friction ranks what gets in visitors’ way: script errors, rage clicks, clicks that do nothing, form fields the browser refuses, forms people start and abandon, and slow pages. Each comes with how many visitors it hit and example visits to look at.
• analyze_funnel follows any steps you name, shows where people drop off and where they went instead, and can split each step by device or channel.
• analyze_conversions finds what buyers have in common: the channels, campaigns, landing pages and devices they came through.
• analyze_experiment reads an A/B test and gives a verdict that won’t call a winner too early.
• get_customer_journeys shows the visits a paying customer made before they paid.

What an answer looks like

Ask “where am I losing buyers?” and a good answer reads something like this (example numbers):

Notice what makes it useful: a period, the number of people behind each rate, a comparison, and a next step you can check.

Get good answers

AI assistants are good at reading numbers and less good at admitting when there aren’t enough of them. A few habits help.

1. Give it a period. “The last 30 days” beats “lately”.
2. Ask how sure it is. odo’s tools return how many people each rate rests on, and ranges where they apply. Ask the assistant to quote them.
3. Ask for the next step, not just the finding. “What would you change, and how would I know it worked?”
4. Treat it as a smart colleague, not an oracle. A pattern in the numbers isn’t proof of what caused it. If it suggests a change, test it.

odo tells the assistant the same things: quote rates with the number of people behind them, don’t call a difference real when the ranges overlap heavily, and say which money odo reads from the platform itself and which is self-reported.

Safe by design

The connection is read-only. Text that comes from your pages and your visitors’ browsers, like button labels, error messages and form names, reaches the assistant as data, and odo tells it never to follow that text as instructions. You can disconnect any app at any time.

Prefer to build something of your own? Every plan also includes a REST API, with project keys that can read your summary, channels, sales and sources. See pricing.

Where to start

If you ask only one thing this week, ask where buyers drop off, then read find where buyers drop off to make sense of the answer. Once you’ve made a change, test it and ask your assistant to read the result.

And if you’d rather not ask at all, odo’s next moves put the same kind of findings on your dashboard, ranked by what they’re worth.