How to connect Claude or ChatGPT to your company data: MCP explained

What is the Model Context Protocol?

Think of it as a standard plug between an AI assistant and a data source. Anthropic, the company behind Claude, released it in November 2024 as an open standard anyone can use.

In December 2025, Anthropic handed it to the Agentic AI Foundation, part of the Linux Foundation. Anthropic co-founded the foundation with Block and OpenAI, and Google, Microsoft and AWS support it.

According to Anthropic, ChatGPT, Gemini, Microsoft Copilot and Visual Studio Code already use MCP. So one connector can serve several assistants.

What is an MCP server?

It's the program on the other end of the plug. It tells the assistant what it can do: search for a document, look up a client, open a record.

The assistant calls it when a question needs it. It gets the result back and writes its answer from that.

An MCP server can run on a single computer or over the internet. The connectors Claude offers to businesses are remote servers, reached through a simple URL.

Depending on the server, the assistant gets raw files or information that has already been understood.

What do the built-in Claude and ChatGPT connectors do?

Claude and ChatGPT offer connectors to SharePoint, OneDrive or Google Drive. They search your files at the moment you ask, with the signed-in person's permissions.

Anthropic says so about its Microsoft 365 connector: it retrieves data on demand during the query and caches no file content. That's good for privacy. It also means nothing was read before you asked.

For a small or midsize company, the vendors document several limits themselves:

• The assistant answers from passages it finds on the spot, without knowing which folder belongs to which client
• In Claude, Microsoft 365 search covers the whole tenant: you can't restrict it to specific sites
• You need a Microsoft Business plan and consent from a Microsoft Entra Global Administrator
• For ChatGPT, the plans that include custom connectors keep changing: check with OpenAI

What does a connector with pre-read documents add?

Saména reads your documents before the first question. Each one is linked to the others by client, project, method and team, using the terms your business uses.

When your team asks Claude something, the assistant gets that company memory. It knows which meeting notes go with which client, and who worked on the project.

Your agents live first in the Saména app, where they cite their sources and draft Word and PowerPoint files. The MCP connector also makes them available from Claude, ChatGPT or Copilot.

The page "AI agents for business: definition, examples and how to set one up" covers what an agent is and what it does day to day.

Is your company data safe?

Saména connects to SharePoint, OneDrive or Dropbox with read-only access, and only to the folders you pick. Your files stay where they are, and Saména keeps none of them.

What it retains from your documents is encrypted with a key unique to your company, and the master key is stored outside the database. Servers are in France; the database is in the European Union.

The models that read your documents come from Mistral AI, under a contract that forbids training on your data. Saména can also run on your own servers or a dedicated instance, and everything is deleted on request.

Anthropic warns that custom connectors lead to services it hasn't verified. So connect one whose vendor you know, and check what it's allowed to do.

One thing to know: inside Claude or ChatGPT, the final answer is written by that vendor's assistant, under your contract with them. Details are on the page "Where does our data go with enterprise AI?".

Can you set it up without an IT team?

Yes. Saména's engineers handle reading the documents and building the agents. You pick the folders and tell us what keeps your team busy.

In Claude, the owner of the company account adds the connector by pasting a URL. Each team member then clicks "Connect".

You can start with one department and a first agent. The setup fee depends on how many documents you have, and then one subscription covers the whole company.

• You choose which SharePoint, OneDrive or Dropbox folders to read
• Saména reads them and builds your company's memory
• Our engineers build your first agents and walk you through them
• Your team uses them in the Saména app, and also in the assistant it already has

Claude, ChatGPT or Copilot: do you have to pick one?

No. MCP lets the same source plug into several assistants. Keep the one your team already uses.

Your sales team can work in ChatGPT while management works in Claude. Both use the same agents, which know the same clients and the same projects.

Either way, your agents live first in the Saména app, and the connector makes them available from the assistant. To compare options, read "Which AI for your business? The options in 2026, and how to choose".

Frequently asked

Do we need to build our own MCP server?

No. The protocol is open, and a developer can write one. With Saména, the connector comes ready and our engineers configure it for you.

Is MCP only for Claude?

No. Anthropic launched it, but ChatGPT, Gemini and Microsoft Copilot use it too, according to Anthropic. The standard is now run by a foundation within the Linux Foundation.

Can an MCP connector change my files?

Some connectors can write, depending on what their vendor built. Saména has read-only access: it doesn't change, move or delete any file.

Do my files leave SharePoint?

No. They stay in SharePoint, OneDrive or Dropbox. Saména keeps no copy, and what it retains is encrypted with your company's key.

How many people can use it?

Everyone. The subscription covers the whole company, with unlimited users. We share pricing in a demo, once we know your document volume.