Home
Services
AI & AutomationWeb Apps & MVPContent CreationPricing
Case Studies
About
AgencyStacksKnowledge
Language
DEEN
Accessibility
Let’s connect

AI & Automation

MCP explained simply: how AI assistants access your software

What the Model Context Protocol is, how it connects AI with your CRM, ERP and more, and what to watch for with permissions and security.

4 September 2026 · Levisteria

Mentioned in this article
Model Context ProtocolAnthropicOpenAIGitHubn8n

MCP, short for Model Context Protocol, is an open standard that lets AI assistants use the tools and data of other programs, for example your CRM, project software or accounting. Anthropic introduced MCP in November 2024. Since December 2025 the standard has belonged to the Agentic AI Foundation under the Linux Foundation, co-founded by Anthropic, Block and OpenAI. When the foundation launched, more than 10,000 MCP servers had already been published.

What problem does MCP solve?

Without a shared standard, every AI assistant needs its own connection to every program. With five assistants and twenty programs, that is a hundred separate integrations that all need maintaining.

With MCP, a program provides a so-called MCP server once. Every assistant that understands MCP can use it, much like a standard plug that fits every matching socket.

How does MCP work?

There are two sides:

  • The MCP server sits with the program and offers tools, such as "search projects", "book time" or "check budget". It checks every request to see whether it is allowed.
  • The MCP client is the AI assistant or an automation, such as Claude, ChatGPT or a workflow in n8n. It asks the server which tools exist and calls them with the right details.

For your team it looks like this: someone types "book two hours on the website relaunch project" in the chat. The assistant finds the project via the MCP server, books the time and confirms.

What does this look like in practice?

We developed our own MCP server for the project software Moco. With it, an AI assistant can search projects, book time and check the budget, with exactly the permissions we grant. We published the code on GitHub.

What should you watch for in terms of security?

An MCP server opens a program up to AI. That is why a few rules apply:

  1. Permissions as narrow as possible: where reading is enough, there is no writing.
  2. Approval for important actions: a person confirms bookings, deletions or payments.
  3. Only trusted servers: third-party MCP servers can inject hidden instructions into the AI. Only use servers whose origin you know.
  4. Store credentials securely: do not keep keys in plain text, and issue access with an expiry date.
  5. Logging: every call is stored and can be traced.

When is a dedicated MCP server worth it?

Ready-made MCP servers already exist for many well-known programs. A dedicated one is worth it if your team works with an AI assistant every day and needs data from a system that has no connection yet, or if you want finer control over what the AI may do. How MCP works together with agents is explained in AI agents for mid-sized companies.

Sources

← All articles

Let’s connect

Should your software talk to AI?

We check whether a connection for your system already exists and otherwise build a dedicated MCP server with clear permissions.

Get your workflow checked →

30 min · free · an honest take on whether it's worth it

Or email ushello@levisteria.com
What happens3 steps
first call — 30 minfree
  1. You tell us where things get stuck.
  2. We ask questions and look at your workflows.
  3. You get an honest assessment and, if it fits, a suggestion for the next step.
→ Then: a concept with clear effort and benefit
Open source tools we use
Open WebUIOllamaMetaMCPOpenCoden8nSupabaseTwentyCal.comChatwootPlaneAFFiNEOpenPanelMetabaseGlitchTipCoolify

Accessibility

Adjust the page. Your choice only applies on this device.

Accessibility statement →