AI Automation

What Is MCP (Model Context Protocol)? In Plain Words

MCP, the Model Context Protocol, is an open standard that lets AI apps connect to your tools and data. A plain explanation with business examples and risks.

  • By
  • Updated
  • 3 min read
  • Facts checked
Short answer

MCP (Model Context Protocol) is an open standard for connecting AI applications such as Claude or ChatGPT to external data and tools, such as files, databases, calendars and business apps. Its own documentation compares it to a USB-C port for AI. It saves custom integration work, but every connection also needs permissions and approval.

Part of the guide: AI Automation in Malaysia: The 2026 Guide for SMEs

On this page
  1. What does MCP do, in one picture?
  2. What can a business do with it?
  3. What is the catch?
  4. Should a Malaysian SME care yet?
  5. How does MCP work underneath?
  6. What does a safe first trial look like?
  7. What should you do next?
  8. Sources
  9. Frequently asked questions

Without a standard, every AI app needs its own custom connection to every system. MCP replaces that with one shared way to describe tools and data, so an assistant that supports MCP can use any MCP server.

Part of the guide: AI automation in Malaysia. Checked 22 October 2026 against modelcontextprotocol.io.

What does MCP do, in one picture?

The MCP documentation describes it as a USB-C port for AI applications: one standard connector instead of a different cable for each device. Three parts:

Part Role Example
Host or client The AI app Claude, ChatGPT, an IDE
MCP server Exposes tools and data A server for your CRM or file store
Tools and resources What can be used "search contacts", "read this folder"

What can a business do with it?

  • Let an AI assistant read project files and a calendar to prepare a meeting brief.
  • Let a support assistant search the knowledge base and the order system through one setup.
  • Let a developer assistant read a design file and generate the page.

The MCP site lists examples such as agents using Google Calendar and Notion, and enterprise chatbots connecting to several databases.

What is the catch?

MCP makes connecting easy, which makes over-connecting easy. A server that can write to your CRM lets the AI write to your CRM. Treat each MCP server like a new employee account: minimum access, a named owner, logging, and approval for writes. Hostile text inside an email or web page can try to steer the assistant (see Prompt injection: the risk for business agents). For the control list, use the AI agent governance checklist. A longer look at what MCP changes for paperwork is in LLM and MCP: what changed.

Should a Malaysian SME care yet?

If you already use an AI assistant daily and lose time copying between apps, yes, try one read-only connection. If you run set processes such as invoices or lead routing, a workflow tool is simpler and more predictable. Compare in Automation or AI first?.

How does MCP work underneath?

The specification describes three roles: hosts are the AI applications that start connections, clients are connectors inside the host, and servers provide context and capabilities. Messages use JSON-RPC 2.0 (MCP specification). The same document sets out safety principles: hosts must obtain explicit user consent before exposing user data to servers, tools represent arbitrary code execution and should be treated with caution, and the interface should let users review and authorise actions. In plain terms, the standard makes connecting easy but leaves the permission decisions to you, which is why the controls above matter.

What does a safe first trial look like?

Pick one assistant and one read-only server, for example a folder of price lists. Give it its own account with access to that folder only, write down who owns it, and ask ten real questions you already know the answers to. If the answers are right and the log shows only reads, you have a working pattern. Add a write permission only after that, and only with a person approving each write.

What should you do next?

Not sure whether MCP, n8n or plain integration suits your case? Book a call and see our software service.

Participants connect AI to their own systems through APIs and MCP in the integration module of our AI automation course.

Sources

Frequently asked questions

Who created MCP?

Anthropic introduced MCP as an open-source standard in November 2024. Since then other AI vendors and developer tools have added support, as the protocol's documentation lists.

Do I need MCP to automate my business?

No. n8n, Make and plain API connections work without it. MCP is useful when you want an AI assistant to use many tools through one standard connector.

Is MCP secure?

The protocol does not make a connection safe by itself. Security depends on what the server exposes, the permissions you grant and whether a person approves risky actions. Prompt injection is a known risk.