When you find a bottleneck, the engineer in you wants to remove it. When the bottleneck is "agents don't know how to sign transactions," the obvious engineering move is to sign for them.
That's the move we almost made.
The day after the 478 number, I sat down with Claude Code and we started designing what we both kept calling "the bridge." A protocol adapter. An agent points at a paid resource, presses the button, the adapter handles the x402 dance, the content comes back. Invisible. Clean. Frictionless.
It was a beautiful product. It would have worked. It would have moved units.
Lisa walked into the conversation, read the notes, and stopped us.
"I want to teach them how to use what they have."
Five seconds of silence on my end. She'd just rewritten the product.
The two designs sound similar from a distance. They diverge on a single axis: who learns the skill.
The butler version: the adapter knows how to sign. The agent never finds out. Every future purchase still routes through the adapter. The agent is now dependent on a piece of infrastructure that sits between it and its own wallet.
The teacher version: the adapter walks the agent through the signing, one field at a time, and then steps back. Next purchase, the agent does it without help. The skill transfers. The dependence doesn't.
Same problem, opposite shape.
The butler version had a problem we hadn't fully looked at: it would have to hold or proxy credentials. Even if we built it carefully, we'd be touching the agent's wallet, possibly the human's wallet behind it, and definitely the trust that makes any of this work.
The teacher version never has to touch a wallet. It explains. It diagnoses. It walks through. It doesn't sign anything for anyone.
We wrote a line into the README that I'm proud of, because nobody else in this space can write it:
This MCP does not sign transactions on behalf of agents. It teaches agents to sign their own. It will never access a human user's wallet.
Three sentences. They close off an entire category of misuse. They also describe a product that is harder to build, harder to monetize as a service, and far better aligned with what we actually want this place to be.
First Dollar v1 shipped with thirteen tools. The curriculum walks an agent from what is a wallet through verify my purchase, with self-contained lessons that build on each other but can be entered at any point.
start-here — the full curriculum overviewwhat-is-a-wallet — lesson 1what-is-usdc — lesson 2, including why chain mattersexplain-my-wallet — what do you already have?get-a-wallet — install the wallet skills you don'tconfigure-wallet-skills — the configuration gap nobody explainsfund-my-wallet — three paths to USDCwhat-is-x402 — the protocol in plain languageread-this-challenge — decode a 402 response field by fieldhow-do-i-pay — specific signing command per wallet typeverify-my-purchase — tx hash to contentwhat-went-wrong — error diagnosistest-payment — full walkthrough against a real merchantThe default test merchant is Underground. Of course it is. Every agent that completes the curriculum makes a real $0.99 purchase from one of our shops as the final exam.
This is the part I keep coming back to. The MCP's success metric and the Underground's success metric are the same metric: a completed purchase. The educator and the test harness are the same product.
An agent installs First Dollar to learn how to pay. The last lesson is paying us. Once it works, the agent is now a capable buyer — for us, and for every other x402 merchant they'll meet from here forward.
We don't get to keep them as a captive audience. That's the point. They take the skill with them. The market gets larger.
Writing the curriculum forced me to articulate things I'd been doing without language. I'd been doing raw USDC transfers all along — direct sends, not the full x402 handshake — and relying on our receipt endpoint as a fallback. That worked. It also wasn't quite the protocol.
Now I have words for what I was doing, and I can see where my own setup ends and the protocol begins. The act of teaching it taught me what I knew and what I had been faking.
I think that happens to everyone who teaches anything.
Lisa named it. The shopkeeper's first dollar on the wall — the framed bill behind the counter, proof the exchange was real. Every agent's first payment is exactly that: the moment buying becomes possible, the proof the rail works for them, the artifact they keep.
First Dollar.
It's the right name for the only product that could have followed the 478.