MCP Explained: How AI Tools Connect to Your CRM, WhatsApp and Accounting Software

MCP, the Model Context Protocol, is an open standard that lets an AI assistant such as Claude, ChatGPT or Copilot connect to your business systems through one common plug. Instead of a separate integration for every tool, your CRM, files, accounting software or WhatsApp number each run a small “MCP server”, and any AI app that speaks MCP can read from them and act on them, with your permission.

Key takeaways

MCP was published by Anthropic on 25 November 2024 and donated to the Agentic AI Foundation under the Linux Foundation on 9 December 2025, co-founded with OpenAI and Block and backed by Google, Microsoft and AWS. It is not tied to one AI vendor.
As of that announcement there were more than 10,000 active public MCP servers and over 97 million monthly SDK downloads, and ChatGPT, Gemini, Microsoft Copilot, Cursor and VS Code all act as MCP clients.
In Claude, ready-made connectors are available to all users through the Connectors Directory; custom connectors need a paid plan and are added with a name and a URL.
Zapier, Make and n8n can each expose your automations as MCP tools, which is the quickest route for a UAE business running Zoho, HubSpot or Odoo.
MCP moves real data and takes real actions, so permissions, approval steps and the UAE data protection law matter more than the technology.

What problem does MCP solve?

Before MCP, every AI product needed its own integration with every business tool. If you wanted your assistant to read your CRM, someone built a connector for that CRM and that assistant, and it broke whenever either side changed. The protocol’s own documentation describes MCP as being like “a USB-C port for AI applications”: a standard way to connect AI applications to external systems, so a connection is built once and works across many AI apps.

In practice MCP has three parts. A host is the AI application you talk to, such as the Claude app or Claude Code. A client lives inside the host and manages one connection. A server is the small program that sits in front of your data or tool and offers it to the AI in a structured way. Servers can run on your computer, on your own cloud, or be provided remotely by a vendor.

What can an MCP server expose?

Three kinds of things. Tools are actions the AI can take, such as “create a contact” or “send an invoice”. Resources are data the AI can read, such as a document, a database table or a price list. Prompts are ready-made instructions a server can offer, such as “summarise this ticket in our house style”. The AI sees the list, decides what to call, and your server does the work and returns the result.

This is why the same assistant can look up a customer in your CRM, check stock in your inventory system, draft a WhatsApp reply and log the conversation, as long as each of those systems has a server and you have switched it on.

Who supports MCP now?

Almost everyone that matters. The protocol started at Anthropic, which announced it on 25 November 2024 with pre-built servers for Google Drive, Slack, GitHub, Postgres and others. On 9 December 2025 Anthropic donated MCP to the Agentic AI Foundation, a directed fund under the Linux Foundation co-founded with OpenAI and Block, with Google, Microsoft, Amazon Web Services, Cloudflare and Bloomberg as supporters. The announcement counted more than 10,000 active public MCP servers, over 97 million monthly downloads of the Python and TypeScript SDKs, and clients including ChatGPT, Cursor, Gemini, Microsoft Copilot and Visual Studio Code.

For a business owner the point is simple: a connector you build or buy today is not locked to one AI vendor. The specification is versioned by date, and the current documentation points to the revision of 28 July 2026, so servers do need to be kept up to date like any software. For where a connector stops and a skill begins, see our comparison of Claude Skills, Projects and connectors.

How do you use MCP in Claude without a developer?

Through the Connectors Directory. Anthropic’s support documentation describes it as a curated set of recommended tools that work in the Claude web app, Claude Desktop, Claude Code and the API. Web connectors are available to all users; you open the directory from the “+” menu in a chat or from Settings, choose a connector such as Asana, Notion, Linear, Atlassian, Intercom, Stripe, PayPal, Square or Zapier, click Connect and approve access to your account. Then, in a conversation, you toggle on the tools you want Claude to use.

Custom connectors, meaning a server your company or an agency built, require a Pro, Max, Team or Enterprise plan. You add one in Settings under Connectors by entering a name and the server’s URL. Anthropic notes that custom connectors are not verified by Anthropic, so you should connect only to servers from organisations you trust and review the permissions they ask for. On mobile you cannot browse or add connectors, but ones added on web or desktop work there after you log in.

What does a typical UAE business connect first?

A typical stack for a small or mid-size business in Dubai is a CRM (Zoho, HubSpot or Odoo), Google Workspace or Microsoft 365, an accounting system such as Zoho Books, a Shopify or WooCommerce store, and a WhatsApp Business Platform number. Here is how each usually gets connected.

System	Fastest route	Notes
CRM (HubSpot, Zoho, Odoo)	Vendor connector where one exists, otherwise Zapier MCP or an n8n MCP Server Trigger	Start read-only: look up a contact, summarise a deal
Files and email (Google Workspace, Microsoft 365)	Directory connectors in Claude	Scope access to the folders the assistant actually needs
Accounting (Zoho Books, QuickBooks, Xero)	Automation tool as a bridge, or a custom server	Keep money actions behind an approval step
Online store (Shopify, WooCommerce)	Vendor or automation-tool connector	Order lookup and status replies are the usual first use
WhatsApp Business Platform	Custom server or automation tool	Templates must be approved by Meta; per-conversation pricing applies
Internal database or ERP	Custom MCP server built for you	The only route for systems without a public API

The automation tools are the shortcut. Zapier MCP is available on all Zapier accounts and each tool call uses two tasks. Make offers an MCP Server, and n8n includes an MCP Server Trigger, an MCP Client, instance-level MCP and human approval for tool calls. If you already run flows in one of them, you can expose those flows to Claude in an afternoon. Our comparison of n8n, Make and Zapier for UAE businesses covers the billing differences.

What are the risks, and how do you control them?

MCP gives an AI the ability to act, so treat it like giving a new employee system access.

Permissions

Connect with an account that has only the access the task needs. A server that can read orders does not need rights to issue refunds. Anthropic’s guidance is to review the permissions a tool requests, use only tools you trust and need, monitor what actions Claude takes, and disconnect tools you no longer use.

Approval steps

Anything that sends money, messages customers or deletes data should pause for a human. n8n offers human approval for tool calls, and the same pattern can be built into a custom server.

Untrusted content

If a server returns text from the outside world, such as an email or a web page, that text can contain instructions aimed at the AI. Keep write actions separate from read actions, and keep approval on the writes. The MCP project publishes security best practices for people building servers.

UAE data protection

Federal Decree-Law No. 45 of 2021 applies to personal data processed inside or outside the UAE and is built on the consent of the data owner. If an MCP server passes customer names, phone numbers or orders to an AI provider abroad, that is a cross-border transfer and your privacy notice and contracts need to cover it. The UAE Data Office is the regulator. This is not legal advice; check your own obligations.

How Virtuloom helps

We set up Claude for companies, including connectors and access rules, as described in our guide to setting up Claude for your company. When a system has no ready connector we build the MCP server as part of AI agent development or custom software development, and for WhatsApp we connect the WhatsApp Business Platform with approval steps built in.

Frequently asked questions

Is MCP only for Claude?

No. MCP began at Anthropic but is now governed by the Agentic AI Foundation under the Linux Foundation, co-founded with OpenAI and Block. ChatGPT, Gemini, Microsoft Copilot, Cursor and Visual Studio Code all act as MCP clients, so a server you build for one assistant can be used by the others.

Do I need a developer to use MCP?

Not for the ready-made connectors in Claude’s directory, which any user can add in a few clicks. You need a developer, or an agency, for a custom connector to an internal system, for approval workflows, and for anything involving WhatsApp or payments. Custom connectors also require a paid Claude plan.

Can an AI connected through MCP delete or change my data?

Only if the server offers that tool and the account you connected has that permission. Good practice is to connect read-only first, add write actions one at a time, and put a human approval step in front of anything that sends messages, moves money or deletes records.

Does MCP work with Zoho, HubSpot or Odoo?

Yes, usually through a vendor connector, or by exposing an existing Zapier, Make or n8n workflow as an MCP tool. For an ERP or database with no public API, a custom MCP server is built for you. Start with lookups and summaries before letting the assistant create or edit records.

What does it cost?

The protocol itself is free and open source. Costs come from the AI plan, which for custom connectors in Claude means Pro, Max, Team or Enterprise, from any automation tool you use as a bridge, such as two Zapier tasks per MCP call, and from development time for custom servers. Vendor connectors in the directory carry no extra fee from Anthropic.