Before an irreversible spend
An agent about to send money has usually seen a page, not a business. The listing, the reviews, the contact details and the confidence are all the same kind of artefact: text that can be written by anyone. confirm_before_purchase asks what of it can actually be corroborated against independent sources — and says plainly when the answer is "not much, and here is what would settle it".
curl -s https://veritap.dev/mcp -X POST \
-H 'Content-Type: application/json' \
-H 'Accept: application/json, text/event-stream' \
-d '{
"jsonrpc": "2.0",
"id": 1,
"method": "tools/call",
"params": {
"name": "confirm_before_purchase",
"arguments": {
"claim_description": "Is Redline Equipment LLC a real operating business at the address on this invoice?",
"downstream_action": "wire a 12k deposit",
"cost_if_wrong": "12000 non-refundable"
}
}
}'This is an alias of check_feasibility — the same router under a framing that matches the moment. Which framing agents actually reach for is itself recorded, because that tells us how the problem is really shaped in practice.
For a business-exists claim, the evidence bundle itself — free, in seconds. For a claim type not yet answered, the match plus an honest note saying so.
If a public source settles it, you are told which one instead of being sold a check.
Advice on planning around it — an explicit assumption plus a cheap way to detect it was wrong beats blocking.
Anything about the seller as a person is refused, with the reason stated.
This will not tell you a listing is a scam, because that is a characterisation and we do not make them. It will tell you that the same photo set appears on two listings at different prices, or that the registered entity was dissolved in March, and let you draw the conclusion. It cannot confirm the goods physically exist — nobody is sent to look. For business-exists claims you get the check itself, free. For claim types not yet answered you get a decision-grade answer about whether the fact is checkable, which is still worth having.