Claude and MCP
Connect Claude to your systems, with permissions per role
Out of the box, Claude knows nothing about your orders, customers or agreements. That is why we connect Claude to your own systems through MCP or an API, such as SharePoint, Exact or RentMagic. Claude then answers with today's data, and only with what the person asking is allowed to see.
- Live data through MCP or an API
- Search your documents on meaning
- Every connection studied carefully first
Claude with your own data
How connecting Claude to your own data works
Anyone who wants to connect Claude to Exact, SharePoint or another system is really asking two things. How does Claude reach what is in that system right now? And how does Claude find the right passage among hundreds of documents? For the first we use MCP or an API, for the second a search index that searches on meaning.
MCP stands for Model Context Protocol: a set way for an AI model to talk to another system. Through such a connection Claude can look up an invoice in Exact, a contact in Business Central or an order in RentMagic, at the very moment someone asks for it.
The connections run in Managed Cowork, the AI workspace we set up and manage for you. There, an extra check sits between your systems and Claude, so the permissions of the person asking always apply.
Live data
Claude MCP: straight from your own systems
A connection through MCP or an API fetches the data while you work. Ask about a customer's outstanding invoices or the status of a ticket in Autotask, and Claude looks in the system itself. So you are always working with current figures.
SharePoint, OneDrive, Outlook and Teams come with Cowork as standard. For your file server, Slack, ClickUp, GitHub and RentMagic the connection is ready to go. An ERP or finance package, or your own software, we connect as custom work.
- Always included: SharePoint, OneDrive, Outlook and Teams
- Ready to use: file server, Slack, ClickUp, GitHub and RentMagic
- Custom work for, say, an ERP or finance package
- A custom connection then becomes part of our management
Documents
Search on meaning, through a dedicated MCP server
For large collections of manuals, PDFs and procedures we build a search index that looks for meaning, not just matching words. The technique is called RAG (retrieval augmented generation). We put the index behind an MCP server, so Claude searches it on its own whenever a question calls for it.
Does the same knowledge also need to work outside Claude, for example as a knowledge base for customers on your website? Then we build a separate RAG pipeline. Together we choose what fits your questions.
- Finds passages even when they use different words
- Claude decides for itself when to search the index
- Answers come with the source document
- We keep the index up to date as documents are added
Permissions
What Claude retrieves depends on who is asking
A connection with one general key can often reach everything in a system. That is why we place our own check between your systems and Claude. There, the permissions of the person asking apply before any data goes back.
Per connection and per role you set what is allowed: nothing, read, write or manage. Write access is off by default, and an AI suggestion is only carried out once a person approves it.
- Permissions per connection and per role
- Write access off by default
- Human approval for every suggestion
- Every change and every answer is logged
Approach
From system to connection in four steps
- 1
Take stock
Which systems and documents are involved, which questions do you want to ask, and who may see what?
- 2
Study
For each system we find out whether a connection through MCP or an API is possible, together with you and the software supplier.
- 3
Build and test
We build the connection and the search index, and test with questions your colleagues really ask, until the answers are right.
- 4
Manage
From then on the connection is part of our management and follows the permissions you have set.
Every connection is subject to study: we examine the software thoroughly first, together with you and the supplier, and only then confirm what is possible.
Frequently asked questions
Questions about connecting Claude
What exactly is MCP?
MCP (Model Context Protocol) is a set way for an AI model such as Claude to request data from another system. We build and manage these connections in Managed Cowork, with an extra check on the permissions of the person asking.
Can Claude access our data in Exact?
Often, yes, but subject to study. Whether a connection with Exact, Business Central or another package is possible depends on the software. We study that thoroughly first, together with you and the software supplier, before we commit to anything.
What is the difference between MCP and RAG?
Through MCP, Claude fetches live data from a system, such as an order or an invoice. RAG is a search index that searches large collections of documents on meaning. We often combine them: we put the search index behind an MCP server, so Claude can search it itself.
Can Claude also change things in our systems?
Only if you allow it. Write access is off by default and you switch it on per connection and per role. Even then, a suggestion is only carried out after a person approves it, and everything is logged.
Does a connection keep working if we switch AI model?
Yes. The connections and the permission check are separate from the model. If you later add ChatGPT or the private model next to Claude, your permissions stay as they are. What we build with AI is regular software that we host and manage for you, and it keeps running without AI.
See Digital Care in your own environment
A half-hour conversation with one of our specialists. Your own systems, your own questions. No sales pitch.