Short version of how purchase_hotel works end to end:

1. Agent calls the tool with a rate ID and guest details.

2. Server returns HTTP 402 with a payment challenge (amount, recipient, expiry).

3. Agent signs a USDC authorization on Base. No wallet UI, no user involvement.

4. Agent retries the same tool call with the signature attached.

5. A facilitator submits the transfer on-chain. It settles in ~2 seconds.

6. The server continues the booking inline and returns a confirmation.

Hotels was an obvious first target for us, but I'm curious what else people would let an agent spend money on without them watching. Flights. Concert tickets. Groceries. A dentist appointment. A specific GPU hour on some obscure cloud.

I built Trip1 MCP and spent the last few months getting x402 to work end to end for hotel bookings. Happy to take questions on any of it.

Things worth asking about:

• How the x402 handshake actually runs inside an MCP tool call

• Why we settled on Base and USDC

• Supplier side: how a real hotel inventory system takes a transfer that wasn't signed by a human

• What breaks when an agent doesn't have a wallet loaded

• Why MCP rather than a REST API with API keys

• What I'd change about the x402 spec if I could

Drop a question, I'll work through them.