An MCP app for Claude, ChatGPT & co.

Your assistant knows products. It does not know prices.

It quotes them anyway. PriceLayer verifies every Amazon ASIN live, returns the price with a timestamp, and a link that actually goes to that product.

Connector URL
https://mcp.pricelayer.net/shopping
  • No account, no token
  • Self-hosted
  • Every price carries its timestamp

The problem

A language model has no prices. It has memories.

Training data is months old; prices change daily. Asked for a price, the model answers anyway — and it sounds convincing.

01

The price is a guess

A number from memory, with no date and no source. It may be months out of date, or it may never have been right.

02

The ASIN is invented

Ten characters that look like a real identifier. It points at a different product — or at nothing.

03

The link goes nowhere

Wrong variant, dead page, or a search page instead of the item. The customer does not land where they were sent.

And none of these three failures is visible in the answer.

The difference

Same question, two answers.

On the left, what an assistant says with no tool. On the right, what it says with PriceLayer.

What does the Acer gaming mouse cost right now?

Without PriceLayer

The Acer gaming mouse costs around €24.99. Here is the link: amazon.de/dp/B08XKM3RT2

  • Price from memory — no date, no source
  • ASIN invented — leads to a different product
  • Nothing signals that either is uncertain
With PriceLayer

€19.98 — Acer Gaming Mouse Wired, 12800 DPI (in stock). As of today, 10:13. Purchase link (affiliate link).

  • Price verified live, with the time it was read
  • ASIN confirmed — B0FCMHJBMW
  • Link goes to exactly this item

The fix

Verify before answering.

PriceLayer plugs into the assistant as a tool. When it names an Amazon product, it asks us — and gets facts instead of recollection.

Every ASIN is confirmed

Does the item exist? Is it the right one? Anything that cannot be confirmed is named, not quietly dropped.

A price with its timestamp

The current price, plus the moment it was read. Never guessed, and never called “live” when it is not.

A link that works

Straight to the verified item, with the right variant. Labelled as an affiliate link — in every answer.

Honest when it cannot

If the source does not answer, the tool says exactly that. A guessed price would be worse than none.

Three steps

Add it once, then forget about it.

  1. 1

    Add the endpoint

    One command in Claude Code, or the URL in your connector settings. No account, no token.

  2. 2

    Ask as you always do

    “What does the mouse cost?” — you never mention a tool. The assistant works out for itself when to verify.

  3. 3

    Get a verified answer

    Price, timestamp and a working link. Plus a warning whenever something could not be confirmed.

Install

One URL as a connector — that is it.

This is how Claude, ChatGPT and other assistants plug in external tools. Authless: no account, no token, no sign-up.

Claude, ChatGPT & others

In your app's settings under “Connectors”, add a custom connector and enter this URL. The exact wording differs by app and version.

https://mcp.pricelayer.net/shopping

Claude Code

Terminal client only: there you can use a command instead of the settings screen.

claude mcp add --transport http pricelayer-shopping https://mcp.pricelayer.net/shopping
Full MCP app reference

See for yourself

Try it without installing anything.

Enter an ASIN and see exactly what the assistant receives. Same check, same source.

Check an ASIN

Why you can trust it

Honesty is not a setting here. It is the construction.

Every price carries its timestamp
We state when the price was read. Nothing is presented as “live” that is not.
Affiliate links are labelled
We earn from qualifying purchases. That is stated on every link — in the assistant's answer and on this site.
No data about you
No accounts, no sessions, no tracking of your query. Reach measurement is self-hosted and cookieless.
Nothing is invented
When something is missing, we say so. No filling the gap with a plausible number.