← All posts
Trust & security · 30 April 2026 · 6 min read

How to safely connect Claude to your live business data (without breaking GDPR)

Plugging an AI assistant into your live business data is now technically easy. Doing it without leaking personal data, breaching GDPR or exposing customer addresses to whoever’s next to take over your laptop — that takes some thought.

Five questions to ask before you connect anything

1. Where does the AI run — and what does it remember?

If you’re using Claude, queries are processed by Anthropic. Anthropic’s commercial T&Cs prohibit them from training on your data. ChatGPT has a similar opt-out for business accounts. Verify the specific terms for the tier you’re using.

2. Where does the data live in transit?

For our Linnworks MCP, every query flows: Claude → MCP server (Cloudflare Worker) → Linnworks API → back the same way. Nothing is stored on our side. The data lives in transit for milliseconds.

3. Who has the bearer token?

Bearer tokens are the keys to your data. If a token leaks, anyone with it can read your data. Don’t share. Rotate if you suspect a leak. Don’t paste tokens into Slack.

4. What permissions does the MCP server actually need?

For our Linnworks MCP, read-only on Orders, Inventory, Customers, Purchase Orders, Refunds and Returns. No write access. The MCP server cannot despatch, cancel, refund, change inventory, or do anything destructive.

5. What does the customer-facing data look like?

Customer names, addresses and emails will pass through the MCP server when Claude needs them. If your privacy policy doesn’t already cover AI processing, update it. We’ve put a draft DPA clause in our privacy policy you can adapt.

What we do at MCP-G

What you should do at your end

Want the full security write-up before you commit? Email hello@mcp-g.com.

Ready to try it?

Two clicks gets you a hosted Linnworks MCP connected to Claude. Cancel anytime.

See the Linnworks MCP →