A developer sells an AI tool with one 5 USDC license instead of per-call fees
Most paid MCP servers bill per call. One builder used the same x402 payment system to sell a single license, and explains how it works and what to guard against.
The short version
- The tool, Cierre de turno, summarizes your GitHub activity into one end-of-day sentence; its MCP version costs 5 USDC once.
- Payment runs on x402 over Base mainnet, then the server issues a license token sent as a plain header on every call.
- Only a hash of each license is stored, so a leaked database can't be turned into working licenses.
- Usage is capped at 200 calls per month per license, so a runaway loop can't drain the service.
- The post gives no sales results, so it shows the method, not that the model earns more.
Why it matters
If you build or buy tools for AI agents, the pricing model decides who carries the risk of a runaway loop. A flat license gives buyers a known cost and sellers fewer billing disputes. This post shows that the same payment plumbing can support either approach.
Going against the per-call grain
MCP, the Model Context Protocol, is the standard plug that lets AI agents use outside tools. A growing number of developers sell access to such tools, and as dev.to reported, nearly all of them charge for each call. They do it with x402, a payment handshake built on the old web status code 402, "payment required." The server asks for money, the client pays, and the request goes through.
One developer chose a different route. His tool, Cierre de turno, looks at your public GitHub activity and produces a single sentence to end your workday. It covers what you finished, what is still open, why it is stuck and where to begin tomorrow. The web version is free, and the version for MCP clients such as Cursor or Claude Code costs 5 USDC, paid once, on the Base network.
Why a daily habit doesn't suit micropayments
USDC is a digital dollar, and Base is the blockchain network where the payment settles. The author's reasoning was about rhythm. People would run this tool a couple of times every day, and at that pace many tiny charges add up to something that feels like a subscription, only more fiddly.
He argues that per-call pricing makes sense when each call produces something with clear standalone value, such as a model's output or data the buyer resells. For a small routine tool, he says, a single license fits the way it is actually used.
How the purchase works
The buy endpoint replies with the standard 402 response and a header describing what to pay, to whom, and in which currency. Any client or agent that understands x402 can read it and pay without a manual. It then retries the request with a payment header attached.
The server hands that payment to a facilitator, a third-party service that checks and completes the transaction on the blockchain. Once the payment clears, the server creates a license token tied to the paying wallet. From then on the buyer adds that token as an ordinary header on every MCP call. There are no accounts, no sign-in flow and no dashboard.
Three safeguards the author recommends
First, store only a fingerprint. The server keeps a one-way hash of each token, not the token. A hash is a scrambled value that can't be reversed, so if the storage leaks, nobody can rebuild working licenses from it.
Second, limit heavy use inside the product. A one-time fee with unlimited calls leaves the seller exposed, so each license gets 200 calls per month. The count is kept in a key-value store that is eventually consistent, meaning updates can lag. A few extra calls may slip through under heavy simultaneous use, which he accepts because the cap only needs to prevent loops, not be exact.
Third, test the payment path with fakes. Before any real money moved, he checked that a bogus payment header gets a clean rejection with a proper challenge, and that a valid license unlocks the tool call while the basic handshake and tool listing stay open to everyone. By the time a real wallet paid, the script had already passed many times.
The facilitator detail that nearly tripped him
The public x402.org facilitator is handy while developing but isn't meant for real mainnet payments. For USDC on Base, the author points to free public options that need no account, naming Dexter as the one he uses. Coinbase's CDP facilitator is another choice if you want sanctions and risk screening on every transaction and don't mind having an account.
His practical tip was to make the facilitator address a setting from the start. That turned a later switch into a very small change.
What to take from it, and what to doubt
This is one developer's account, and it includes no sales figures. We don't know whether a license earns more or less than per-call fees would have. What it does show is that the x402 handshake isn't tied to charging per request: one payment can yield one token that keeps working.
For buyers, the appeal is predictability, especially when an agent could loop and make hundreds of calls unattended. For sellers, the trade is less revenue from heavy users in exchange for simpler billing and fewer complaints. The code is public, so anyone can inspect how it was built.
What to watch
- Whether the author or others publish real sales numbers comparing licenses with per-call pricing.
- Whether MCP clients and agents get easier ways to handle license headers and x402 payments automatically.
Try it
If you plan to sell a tool to agents, write down how often a typical buyer would call it each day, then decide whether per-call fees or a capped license fits, and state the choice on your page.
Sources and links
- Reportingdev.to: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by dev.to · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
