INTEGRATIONS · JEWELRY · The Edge + Claude

Does The Edge integrate with Claude?

Not natively — as of August 2026 we know of no vendor-supplied connector from The Edge to Claude or any other AI assistant. It does connect in practice: a nightly read-only export from The Edge lands in a hosted copy, and Claude reads that copy through an MCP gateway. That connection is in production today.

Last verified

What each system does

The Edge is a point-of-sale and store-management system for independent jewelers, from Abbott Jewelry Systems. It runs on a machine in the store and holds the store’s inventory, customers, sales history, repairs and custom orders, and appraisals. It is where the store’s day actually happens.

Claude is the AI assistant from Anthropic. On its own it knows nothing about a particular store. It can be given connectors — over MCP, the Model Context Protocol — that let it look things up in a business’s systems while a person is asking questions.

What connects natively

Nothing, as far as we can find. As of August 2026 The Edge publishes no MCP connector and no public API intended for outside programs; its integrations are vendor-arranged. If that changes, the vendor’s own path would be worth evaluating first, and we will update this page.

What The Edge does have is an export. That is the hinge the whole connection turns on.

What a connection looks like in practice

This is the shape of the deployment that is in production today, since go-live.

  1. The export. Once a night, a little after seven in the morning Eastern, The Edge’s read-only export runs on the store’s machine and lands, outbound over a private link, in a copy we host. The export has run every night since go-live. The POS is never exposed to the AI or to the app; both only ever see the copy.

  2. The gateway. On the AI side there is one MCP gateway. Behind it, at this deployment, are seventeen tools. All seventeen are reads — past-due repairs, customer history, what sold, what’s aging in the case. One of them is a free-form query for the questions nobody anticipated, and it is forced read-only at the server, with a timeout and a row cap, so a clever question cannot become a slow one or a destructive one.

  3. The people. The store runs three permission tiers — owner, manager, associate. Each person adds one connector to their own Claude and carries their own key. What they can ask is decided by the key, not by the phrasing: an associate’s list of tools is shorter, and topics the owner designates are filtered out before a question is even matched, so there is nothing to argue with and nothing to leak in a refusal.

  4. The questions. “Which repairs are past their promised date?” “Who bought bridal last year and hasn’t been in since?” “What did this customer buy the last three times?” All reads. Nobody’s first question is “please change something.”

Limitations and honest caveats

  • It is as current as last night. Anything that happened today is not there until tomorrow morning. If the question is about right now, this connection is the wrong tool.
  • It cannot write. That is the design, and it is still the design. A store that wants writes would need a different path, and a decision to build one. Read-only first: why we sequence access this way →
  • The export is what the export is. The Edge decides what its export contains. Fields it leaves out are not available to Claude, and we don’t reach around it.
  • Someone runs the copy. The export job, the hosted copy, and the gateway are running services. At this deployment that someone is us.
  • We don’t publish a speed figure. We haven’t measured how fast it answers.

Bigger picture

This page is the worked example for two broader questions that any owner-operated business runs into, not just jewelers: What is an MCP connector? → and How do I connect AI to on-premise software? →

Setup guide: Can my AI work with The Edge jewelry POS? →

Who this is for

An independent jeweler on The Edge who already pays for Claude (or is about to), wants to ask plain-language questions about their own store without building reports, and is comfortable with answers as of last night. If that is you, tell us about the store. Read-only first. Writing only when you say so.

QUESTIONS

Questions owners ask

Can Claude see my Edge data in real time?
No. The connection is fed by a nightly export, so Claude answers as of the last run — yesterday's close and everything before it. That covers the questions owners actually bring (past-due repairs, customers gone quiet, what sold last year); it does not cover 'is this piece in the case right now.'
Can Claude change anything in The Edge?
No. There is no path back. The export is one-way, every tool on the connection is a read, and the free-form query tool is forced read-only at the server. Writing to The Edge is not something this connection can do, on purpose.
Does my POS get exposed to the internet?
No. The Edge stays on the store's machine. The export runs outbound over a private link into a copy we host; neither Claude nor the app built on it ever reaches the POS.
Can my associates use it without seeing what the owner sees?
Yes. The deployment runs three permission tiers — owner, manager, associate — and each person's Claude carries that person's key. An associate's assistant has a shorter list of tools; the others aren't refused, they simply aren't there. The owner can also designate topics that lower tiers never reach.
Do I need a special version of Claude?
No. It works from the Claude the person already uses, by adding one connector to it. The gateway does the rest.
What about The Edge and ChatGPT, or other POS systems?
The gateway is built on MCP, which ChatGPT also speaks, but we publish only what we have run in production, and The Edge + Claude is the pairing we have. Other pairings are untested by us until we say otherwise.

If you run The Edge and want to know what your own AI could read from it, tell us about your store. Read-only first. Writing only when you say so.

Tell us what you run