MCP servers
An MCP server is a small, controlled connection that lets your AI tool or your agents read from your own systems and, where you allow it, write drafts into them. It runs with its own access, which you control. We build MCP servers to your warehouse, your ERP and accounting system and your documents, and they run in your own cloud.
This is for you if
- Your people use Claude or another AI tool, and it cannot see your own figures.
- A question about costs waits until someone exports from each system and builds a spreadsheet.
- You want AI on your data, and you want to decide exactly what it may read and change.
What you leave with
- The result
- Your AI tool can answer from your own data. An employee asks a question in the AI tool they already use, and the answer comes with the figures and the records behind it. Each person signs in as themselves and sees only what they may see.
- Read-only first
- The server starts by reading only. Writing comes later, one tool at a time, and only as drafts a person approves in the system they already use.
- Every call logged
- The server writes every call to the logbook: who asked, which tool, what it returned, and when. If the entry cannot be written, no answer goes out.
- Built once
- MCP is an open standard. Claude Team and Enterprise add the server as a custom connector, and other AI applications that speak MCP, such as ChatGPT, can reach it too.
What we do, what you do
| We | You |
|---|---|
| Build the server with only the tools the job needs, and give it only the access those tools need. | Decide which systems it may read, and which tools may write drafts. |
| Run it in your own cloud, behind sign-in per person, with every call written to the logbook. | Control access to the systems and to the server. |
| Set the tool permissions in your AI tool, for example reading allowed and writing blocked. | Add the server once as a connector; in Claude Team, an owner of the Claude account does that. |
| Hand over the code and the settings. Gathena's agents write and maintain the servers, and every change waits for your approval. | Own all of it, in your own cloud. |
Worth knowing
MCP itself does not enforce security; the standard leaves that to the servers and the applications that use it. The limit that holds whatever the AI asks is the server's own access, set in your cloud. The tool permissions in your AI tool are a second check on top. That is why we start read-only and give each server only the access its tools need.
Your questions
- Can the AI change our data?
- Only where you allow it, and only as drafts. A server that reads cannot change anything. A server that writes creates drafts that a person approves in the system they already use, such as a draft entry in the accounting system that the server can never post.
- Which systems can it connect to?
- Your warehouse, your ERP and accounting system, and your documents. Some systems come with their own MCP server; Business Central has one, read-only by default, and it works with Claude.
- Do we need a warehouse first?
- For questions across your figures, yes: the server reads the warehouse, so every answer rests on the same data as your reports. For one system with its own MCP server, you can start with that system on its own.
- Who owns it?
- You. The server, its code and its logbook run in your own cloud, and you control access. The answer to each question passes through the AI tool your people use, at the provider you chose.