How it works
Four steps, and the
third one is the product.
You have already connected things to your agent. What happens next is where every setup quietly stops working — so here is exactly what Customer Magic adds, in order.
- 01
Plug it into the agent you already use
One connection over MCP — the open standard your assistant already speaks. Claude Code, the Claude desktop app, Cursor, or the agent built into the product if you'd rather stay in a browser.
No migration, no import, no CSV, and nothing to tidy first. The sources stay where they are; what changes is that their contents now land somewhere that holds them.
- 02
It sweeps what you've already connected
Inbox, calendar, CRM, billing, product data. Every relationship in there gets read for what it actually is now: who paid, who went quiet, who renews in six weeks, who replied once in 2024 and meant it.
This is the part your agent can already half-do. Ask it who to contact and you get a large, correct, undifferentiated pile — which is the right answer to the question and no use at all.
- 03
Everything sorts into six plays, scored
Retention, expansion, reactivation, nurture, new business, advocacy — ordered by how close the money already is. Each opportunity gets a score and, more importantly, the reason for it.
The reasoning is shown rather than hidden, so you can disagree with it. A renewal at risk never sits behind cold prospecting again, and you can see why it doesn't.
- 04
The first move comes drafted
The research done, the history carried, the message written in your voice, and the smallest physical first step ready to approve. You decide; it does the labour.
This is where the hour actually goes — the gap between “you should contact Marta” and something Marta will reply to. Advice is free and worth about that.
The missing furniture, named
Open your agent as an engineer and it already knows where it is — the repo, the tests, the history, the last four things you tried. Open the same model to do revenue work and you get a text box. That asymmetry isn't about sophistication; engineering's context happens to live in files, and nobody built the equivalent for sales.
Somewhere the result lives
A chat log is not a pipeline. Yesterday's conclusion is still there this morning — including the three you consciously looked at and skipped, which don't come back looking urgent.
An opinion about order
Twenty ranked opportunities is still twenty decisions. One next action is a decision somebody already made. That difference is the whole distance between a tool you open and a tool you use.
Carry-forward
Yesterday's untouched opportunity doesn't restart at zero. The board knows what moved, what stalled, and what you've been avoiding — which is the part no amount of prompting reproduces.
The two-minute test
- 01Ask your agent who you should contact this week. Read the answer — it will probably be good.
- 02Close the session.
- 03Tomorrow, in a fresh session, ask the identical question.
If the second answer doesn't know you dismissed three of the first five, doesn't know you emailed one of them yesterday, and can't say what changed overnight — nothing is holding your work. You are the only persistence layer in the system.
That is not a prompting problem, and no better model fixes it. The long version.
An hour a day. Every opportunity worked.
7 days free, no card. The clock starts the day you open it — not the day you sign up.