Is an MCP CRM too technical for you?

You already talk to ChatGPT or Claude most days. You ask it to draft a reply, tidy up a message, make sense of a supplier email. Then someone mentions MCP and connecting your client file to your AI, and the shutters come down, because that sounds like the part where you need a developer. It is not that part. If you can sign into an app, you can do this, and it takes about as long as making a coffee. This post is here to take the fear out of it.

What is an MCP CRM, in plain words?

Start with the two words behind the acronym. A CRM is just your client file: who your customers are, what happened with each one, what you promised and by when. MCP is the doorway that lets the assistant you already use open that file, read from it, and write to it, without you copying anything back and forth. That is the whole idea. One web address stands for your file, you hand it to your AI once, and from then on the two are talking to each other. If you want the fuller definition of the category, we keep a plain-language glossary entry on the MCP CRM; here we care about one thing, which is getting you connected without stress.

What does the setup actually involve?

Less than you fear. There is no server to run and nothing to install beyond the AI app you already have open in front of you. Start to finish, this is the whole thing:

1. Create your Historis account; it takes a minute and asks for no card.
2. Copy the single connection address Historis shows you, https://api.historis.app/mcp.
3. In Claude, ChatGPT, or Cursor, find where you add a connector and paste that address.
4. Sign in with your Historis account when your assistant asks, and approve the connection.
5. Choose what it may do, from reading only to reading and writing.
6. Ask it something real, such as what is overdue this week.

The exact label on the button moves around a little between assistants, so if you want the click-by-click version for yours, the setup instructions in the docs walk through Claude, ChatGPT, and Cursor one at a time. Read past the six lines above and you will notice what is missing: no code to write, and nothing that belongs on a developer's desk.

What does approving the connection mean?

When you connect, one screen appears, and it is worth reading rather than clicking straight past. You sign in with your Historis account, the same login you use to open the app, and you see in plain terms what the assistant is being allowed to do. To begin with it can read your client file and change nothing at all. Writing is a separate permission you grant on purpose, and you can leave an assistant on read-only for as long as you want to. Nothing here is buried in fine print. The entire point of the screen is that the decision stays yours.

What can the assistant actually see?

Only what you could see yourself, and nothing wider. Historis keeps each note at a visibility you choose: some are open to your whole shop, some sit with one team, and some stay private to you. When your assistant reads through the connection, it is held to those same lines, so a private note stays private and a team's notes stay with that team. Connecting an AI does not quietly open a back door to the rest of your file. It hands the assistant the same view you already have, and no more than that.

Is it safe to let an AI touch your client file?

Sooner or later you will want the assistant to write and not only read, because that is where the time comes back into your day. The worry underneath is fair: what if it writes the wrong thing on the wrong customer? Two things make this calmer than it first sounds. Every change an assistant makes is marked as its own work and kept in the record, so you can always tell your notes from its notes. And an edit it gets wrong can be reversed, because the change is logged rather than quietly applied. If you want the full mechanics, we wrote why letting an AI write is safe about exactly this. It helps too that the AI is yours and not ours: Historis runs no model of its own and only holds your file for the assistant you connect, which is the case we make in bring your own AI.

What changes on the first day?

The morning after you connect, the shift is that you can ask instead of dig. Standing at the counter, you ask your assistant what is overdue this week, and it reads the answer off your shared timeline while you keep serving. Before a regular comes back, you say catch me up on Madame Rossi, and it hands you her history: what she bought, what you promised, what is still open. After a call, you tell it to log the call and follow up on Friday, and the note lands on the right person without you opening a single form. You can still write notes yourself whenever you prefer, since the assistant is an extra pair of hands rather than a replacement for your own. The file stops being something you maintain and becomes something you talk to. That one shift, from filing to asking, is what people mean when they say their AI finally knows the business.

So, do you need to be technical?

No, and it is worth saying plainly. The whole job is pasting one address and reading one screen, and after that your AI and your client file are looking at the same history. If you are still keeping customers in a spreadsheet, connecting an AI is the second move rather than the first: begin by getting off the spreadsheet, then come back and point your assistant at what you built. The belief that this is developer work is the only difficult part, and it happens to be the part that is not true.