MCP × x402 · decision receipts

Auditable receipts for AI shopping agents.

Brands pay to be considered. You get the fee. And a receipt anyone can check.

Live developer preview. Free, no account needed.

receipt.log — shortlisted
$
The shift

The rails are done. The trust is missing.

73%
Consumers who use AI to research purchases.
13%
Have let an agent actually complete an order.
60%
Would drop an agent after a single mistake.
The catch

If a brand pays your agent, whose side is it on?

A brand's fee should buy a seat at the table, not the final answer. We use a flat toll over x402 to enter the shortlist. The fee is stripped from the data before scoring. Plain arithmetic ranks everything.

MCP first

A receipt you can argue with.

Shortlisted is a tool your agent calls before it buys, built as an MCP server. Add it once to Claude or any MCP client, and every shopping decision comes back with cryptographic proof.

  • One MCP server. Add it once to Claude or any MCP client.
  • Every shopping decision returns a signed receipt.
  • Re-runnable arithmetic. Verifiable by anyone.
DECISION RECEIPTrct_8Kx2mN4p
QUERY
wireless earbuds under $100
good battery life
CRITERIAWEIGHT
price0.30<
battery_hours0.30<
sound_quality0.25<
comfort0.15<
SHORTLISTSCORE
NovaSound Air8.42<
Kestrel One8.11<
AcmeBuds Pro7.94<
Tempo Lite7.20<
PAID TO BE CONSIDERED
AcmeBuds$0.02<
Kestrel$0.02<
affected ranking:NO<
COUNTERFACTUAL
Kestrel One wins below $82.40
signature● VALID
Try it

Add it in one line.

01Connect

In Claude Code, run:

claude mcp add --transport http shortlisted https://app.shortlisted.tech/mcp

Using Claude Desktop or claude.ai? Add a custom connector with this URL: https://app.shortlisted.tech/mcp

02Ask

Ask your agent:

Find wireless earbuds under $100 with good battery life. Use Shortlisted and choose the weights yourself.

Your own Claude picks the criteria. Shortlisted ranks with plain arithmetic and signs the result.

03Verify

Then ask: "Verify that receipt." Or paste it into the public verifier. Change any number and watch it fail.

Integrity

Arithmetic, not AI.

The scoring isn't AI.

A model turns your request into criteria, then gets out of the way. The ranking is plain arithmetic. Same inputs, same answer, every time.

Fees can't buy the answer.

The toll is stripped from the data before the scoring code sees it. There's a test proving two identical offers with different fees score the same.

See the test →
Verification needs no account.

Every receipt is signed. Anyone can check one holding only the receipt and our public key. If checking us required our permission, it wouldn't be an audit.

For brands

The toll belongs to you.

Advertising is a toll on attention. Agents remove the attention. The toll gets rebuilt on the shortlist, and it should be paid to you. Join the waitlist for early access.

Brands: tenders and x402 tolls are in preview. Leave your email for early access.