What Is MCP? Using Claude with Your CRM, E-commerce Store and Ad Accounts (2026)

The question "what is MCP?" has moved well beyond software teams over the past year. We no longer ask Claude only general questions. We ask things like "Which product sold best last month?" or "What did we last quote this customer?", questions that depend on your company's own data. Those answers don't live inside the AI. They sit in your CRM, your Drive folders or your store dashboard. MCP is the bridge between the two.

In this article we explain MCP without drowning you in jargon. Then we give concrete answers to the question of what you can actually do with Claude: Gmail and calendar connections, customer lookups in your CRM, order and stock reports, ad and SEO data analysis, and drafting quotes. We work through examples from a furniture manufacturer and a hotel, then cover security, data protection rules and commissioning a custom MCP server for local software that has no ready-made connector.

The goal: by the end, you should see clearly where AI integration can start in your business and where you will need specialist help.

What is MCP and how does it work?

Not long ago, every device had its own charging cable. Then USB-C arrived and one cable started charging phones, laptops and headphones alike. MCP does the same job in the AI world. Announced by Anthropic as an open standard in late 2024, the protocol means you use one shared "plug" instead of writing a separate integration for every piece of software. The official documentation is at modelcontextprotocol.io.

The setup has three parts:

• Client: the AI app you use, for example Claude on the web, desktop or mobile.
• MCP server: a small piece of middleware that sits in front of the system you want to connect. Think of it as the "interpreter" for your CRM, store or Drive account.
• Data source: where the actual data lives. Customer records, orders, ad accounts, emails.

What is an MCP server and what does it do?

An MCP server offers the AI a menu that says "these are the tools I have". A CRM server, for example, might offer: search customers, fetch recent calls, list open quotes. When you ask Claude "Where do we stand with Mr Ahmet's company?", Claude picks the right tool from that menu, the server pulls the data from the CRM, and Claude writes you a readable summary. The server cannot do anything that isn't on the menu, which draws an important line from a security point of view.

Ready-made Claude connectors and how to set them up

Claude connectors are integrations built by Anthropic and its partners that you can add in a few clicks. On the Google Workspace side, Gmail, Google Drive and Google Calendar stand out. The connector directory also includes options for project management, note-taking, design, payments and some CRM platforms. The list keeps growing, so the best place to check what's available is the connector directory in Claude's settings.

Setup steps

1. In your Claude account, open Settings > Connectors.
2. Choose the service you want to connect from the directory (for example Google Drive).
3. In the window that opens, sign in to the relevant account, read which permissions you are granting and approve them.
4. In the chat screen, check in the tools menu that the connector is switched on.
5. Test it with a simple request: "Find the 2025 price list file in Drive and summarise the main items."

Some connectors, and the option to add your own server address (a custom connector), may only be available on paid plans. Plan features are updated often, so check the official page before deciding.

What can you do with Claude? Practical examples from everyday work

Let's leave the theory and look at a typical week in a small or mid-sized business. The table below shows realistic tasks you can give Claude, by connection type.

Looking up customers in your CRM

Where sales teams lose the most time is pulling together a customer's history before a call. If your CRM is connected via MCP, you can simply say: "Summarise Demir Yapı's orders over the last six months, their open quotes and any complaint records, ahead of tomorrow's meeting." If CRM is new to you, start with our article what is a CRM and what it does for a business.

Order and stock reports in e-commerce

Instead of fiddling with dashboard filters on Monday morning, you can tell Claude: "Group the weekend's orders by shipping status and show the ones with return requests separately." Some e-commerce platforms have started publishing their own MCP servers. For local platforms (in Türkiye, for example, many stores run on domestic software), you usually need a custom server built on top of the platform's API.

Analysing ad and SEO data

You can already export ad and analytics data and upload it to Claude. MCP removes that step: the assistant queries the data directly. Requests such as "Compare spend and conversions by campaign and suggest underperforming ad groups" can turn a weekly report into a few minutes' work. Just remember: Claude makes suggestions, you make the budget decisions.

Drafting quotes

When the price list in Drive, the customer details in your CRM and the request email in Gmail come together, Claude can produce a first draft of the quote. You check it and send it. If you want to systemise your quoting process end to end, it makes sense to think about quoting software and AI together.

MCP use cases by sector

Furniture manufacturer (İnegöl/Bursa)

İnegöl, near Bursa, is one of Türkiye's main furniture-making hubs. A manufacturer's day there revolves around dealer orders, fabric and foam stock, production planning and export correspondence. With Claude connected, requests like these become possible:

• "List the 10 dealers who ordered the most this month and any delayed deliveries."
• "Which fabric colours for the corner sofa model are at critical stock levels?"
• "Draft an English reply to the buyer in Germany using the dimensions and prices from the catalogue."

In this sector, most data usually sits in local ERP systems or custom-built dealer systems. If you are thinking of digitising the dealer side, our B2B dealer portal guide is a good starting point. You can see our work with furniture brands on our furniture sector page.

Hotel and tourism business

In hotels, data is spread across the booking system, channel manager, guest emails and review platforms. A Claude instance connected through MCP can handle tasks like these for the front office or sales manager:

• "Summarise occupancy for the next two weeks by room type."
• "Pull out the most frequently repeated complaints in guest reviews from the past month."
• "Write a draft quote for the agency requesting a group booking, with available dates and rates."

The distinction needs to be clear here. With MCP, Claude is an internal assistant for the hotel team. An assistant that talks to guests on your website and guides them to book is a different need. That's where solutions such as OSPITA AI, which talks to guests in 8 languages, come in.

Security, data permissions and data protection

Connecting AI to customer data is exciting, but done carelessly it is risky. MCP itself is a transport standard. What determines security is how the server is written and which permissions you grant.

Practical security rules

• Least privilege: if read access is enough for a report, don't grant write or delete access.
• Approval step: for actions like updating records or sending emails, require Claude to ask for confirmation first.
• User-based access: the sales team's assistant shouldn't see accounting data. The server should filter data based on who is asking.
• Logging: record which tool was called, when and by whom.
• Trusted sources: don't connect random MCP servers you found online to company accounts. A server whose code hasn't been reviewed could send your data to places you don't expect.

Data protection rules (KVKK, GDPR and others)

When personal data such as customer names, phone numbers or addresses is sent to an AI service, that counts as data processing. If the provider's servers are abroad, rules on international data transfers come into play. Rules differ by country: if you operate in Türkiye, this falls under KVKK (the Turkish Personal Data Protection Law); in the UK and EU, GDPR applies. Your privacy notices, records of processing and contracts with the service provider should cover this use.

As a practical measure, you can mask personal data on the server side. In sales analysis, for example, sending a customer code instead of a customer name covers most reporting needs.

Custom MCP servers for local software without a ready-made connector

Many small and mid-sized businesses, especially in Türkiye, run on domestic ERP or accounting software, a CRM they had built for them, or a dealer portal. Most of these systems have no ready-made Claude connector. The good news: if your software has an API, or secure access to its database can be arranged, a custom MCP server can be written for it.

How does the process work?

1. Define the needs: list the questions you'll ask Claude, such as "dealer balance", "delayed orders" or "product stock status".
2. Access analysis: does the existing software have an API, or will a read-only view be created from the database?
3. Tool design: each question becomes a "tool". Permission and filtering rules are defined here.
4. Development and testing: the server is built, tried with test data, and wrong or incomplete answers are fixed.
5. Launch and monitoring: the server goes live in a secure environment, is added to Claude as a custom connector, and usage is logged.

How much does a custom MCP server cost?

The main cost drivers are the quality of the target system's API, the number of tools, whether it's read-only or includes write actions, the need for user-based permissions, and your hosting choice. As a rough guide from the Turkish market: a simple server with a few read tools on a well-documented API may stay in the range of a few tens of thousands of TL, while a server with permissions and write actions for an old ERP with no API can reach hundreds of thousands of TL. These ranges are not fixed prices; they vary significantly with project scope.

For a closer look at how pricing works, see our article on what drives custom software costs. At Fietto Creative we build custom CRM, quoting and dealer systems, so adding an MCP layer to them is a natural part of our custom software service.

Start small: one system, three or four questions, read-only access. Expand the scope as you see value.

Where to start: a step-by-step roadmap

• Week 1: try the Gmail, Drive and Calendar connectors with a test account. Show the team a few example prompts.
• Weeks 2-3: identify the report or query that eats up the most time. Sales follow-up, stock or ad reporting?
• Month 1: check whether a ready-made connector exists for that area. If not, find out the API status of your current software.
• After that: have the data protection side reviewed and, if needed, get a quote for a custom server.

MCP turns AI from a general chat tool into an assistant that knows your business. Set up with the right permissions and in small steps, it lightens the routine load on sales, operations and marketing teams. If you'd like to discuss how it could work with your own systems, get in touch with us.

Frequently asked questions