AI assistants like Claude or ChatGPT are impressive – but out of the box they know nothing about your customer database, your ticketing system or your internal documentation. This is exactly the gap the Model Context Protocol (MCP) closes: an open standard that securely connects AI assistants to your own data, systems and tools.

In this article we explain what MCP servers are, what they are useful for, which FireStorm server is suitable for running them, and how to install your own MCP server and connect it to an AI assistant.

Your own MCP server securely connects AI assistants to your data

Discover our servers

What is the Model Context Protocol (MCP)?

The Model Context Protocol was introduced by Anthropic at the end of 2024 as an open standard and has since become established across the industry – OpenAI, Google and Microsoft now support the protocol as well. Think of MCP as a USB-C port for artificial intelligence: a uniform interface through which any AI application can communicate with any data source or service.

An MCP server is a small program that can provide the AI assistant with three things:

• Tools – actions the AI is allowed to perform, such as a database query, creating a ticket or sending an e-mail
• Resources – data and documents the AI can read, for example product catalogues, manuals or reports
• Prompts – predefined instructions for recurring tasks

The AI assistant (the so-called MCP client) connects to the server, sees which tools are available and uses them autonomously to complete tasks – always within the permissions you have defined.

What are MCP servers useful for?

The possible applications are diverse, ranging from small automation helpers to company-wide AI integration:

• Querying company knowledge: The AI searches internal documentation, wikis or contracts and answers employees’ questions directly – without copying data to third parties.
• Connecting databases: Revenue figures, stock levels or customer data can be queried in natural language (“How many open orders do we have this week?”).
• Automating support: The assistant reads tickets, suggests replies or creates tickets in the helpdesk system itself.
• E-commerce and CRM: Maintain product data, check orders, summarise customer history – all through a single chat interface.
• Development and operations: Trigger deployments, analyse log files or evaluate monitoring data, directly from the AI tool.

The big advantage: instead of programming a separate integration for each application, the MCP server is set up once and then works with all compatible AI clients – such as Claude on the web and as a desktop app, Claude Code, Microsoft Copilot Studio or development environments like VS Code and Cursor.

Why your own server in Switzerland?

As soon as an MCP server accesses real company data, questions of data protection and control arise. If you run the server on your own infrastructure, you retain sovereignty over your data: access can be logged, permissions can be controlled granularly, and the data stays in Switzerland – an important point for anyone subject to the revised Swiss Data Protection Act (revFADP).

Your own vServer also offers full flexibility: you decide which tools the AI may use, you can run several MCP servers in parallel and expand the server at any time.

Which FireStorm server is suitable?

MCP servers are frugal – they merely mediate between the AI client and the data source; the actual computing is done by the AI model at the provider. For getting started, we recommend:

• vServer Green (CHF 10.90/month) – 4 GB RAM, 4 dedicated CPU cores, 80 GB SSD: ideal for first experiments and individual MCP servers, for example connecting a database or document collection.
• vServer Blue (CHF 21.90/month) – 8 GB RAM, 6 CPU cores, 150 GB SSD: our recommendation for production use, when several MCP servers, a search index or additional services are to run on the same system.

Both servers are located in our Swiss data centre, run on 100% renewable energy and include daily backups, DDoS protection and unlimited traffic. As the operating system, we recommend Ubuntu 24.04 LTS or Debian 12.

Via MCP, AI assistants access tools and data sources in a controlled way

Installing an MCP server – step by step

Most MCP servers are based on Node.js or Python and are ready to use in a few minutes. Here is how to proceed on a freshly installed Ubuntu server:

1. Install Node.js

curl -fsSL https://deb.nodesource.com/setup_22.x | bash -
apt install -y nodejs

2. Set up an MCP server

Ready-made MCP servers already exist for many systems – from databases to GitHub to file systems. One example: the official filesystem server makes a defined folder of documents available to the AI. Using a gateway such as Supergateway, it becomes reachable via HTTP:

npx -y supergateway \
  --stdio "npx -y @modelcontextprotocol/server-filesystem /srv/documents" \
  --port 8000

You can develop your own MCP servers with the official SDK (TypeScript or Python) – allowing you to connect any internal system with just a few lines of code.

3. Secure it with SSL

To ensure encrypted and authenticated access, the MCP server is placed behind a reverse proxy (e.g. Nginx) with a Let’s Encrypt certificate. Access should additionally be protected with an access token or OAuth and restricted to the essentials via the firewall.

Connecting the MCP server to the AI

Once the server is reachable, the connection only takes a few clicks:

• Claude (web and desktop): Under Settings → Connectors → Add custom connector, enter the URL of your MCP server – done. The available tools then appear directly in the chat.
• Claude Code (terminal): Connect with a single command: claude mcp add --transport http my-server https://mcp.yourdomain.ch/mcp
• Other clients: VS Code, Cursor and other MCP-compatible applications can also be connected via the server URL.

You can then give the AI tasks such as “Summarise this week’s complaints” – and the assistant autonomously uses the approved tools of your MCP server.

Conclusion: the easiest way to your own AI integration

Within a very short time, MCP has become the standard for connecting AI assistants to external systems. With your own MCP server on Swiss infrastructure, you combine the strengths of modern AI models with full data control – starting at just CHF 10.90 per month.

Our support team will be happy to help you choose the right server and set it up. You can find all vServer offers at www.firestorm.ch/en/rootserver.

Order MCP server