Agents need payment channels, and we've built them three times already

Crypto and agents need each other, and I think payment channels are how they meet. An agent that calls a search API 400 times in a task shouldn't send 400 transactions to a blockchain. It should hand the tool a signed promise each time, and the tool should cash it in when it decides the bill is big enough.

I spent twelve years at CryptoCompare watching people try to make small payments work on chains, and I've seen so many attempts. Most of them had the right idea and the wrong moment. Agents might be the right moment, and it would be nice to combine the best of them instead of starting again.

What would good agent payments look like?

We need something like the Bitcoin Lightning Network, where agents exchange signed transactions, and as the bill grows the provider of the service can execute the signed transaction on chain. This way we reduce noise and we only settle when we need to. Lightning had a cool design, and we should use something like that (or even that) for agent payments without clogging the chain.

Signed transactions that only get settled when the service provider or tool decides to. The agent doesn't wait for a block on every call, the chain doesn't carry a transaction for every search, and the tool holds a signature it can redeem whenever it likes. We probably don't need MCP for any of this either. HTTP already has a status code for it, 402 Payment Required.

Haven't people tried this before?

Many times. In April 2013 Mike Hearn shared Satoshi's description of payment channels, from a private email, with Bitcoin's developers, Jeremy Spilman posted a design for them a few days later, and Hearn and Matt Corallo shipped micropayment channels in bitcoinj in June ^[2]. Joseph Poon and Thaddeus Dryja published the Lightning Network paper in February 2015, and it went live on Bitcoin in 2018 ^[1]. Ethereum got its own version, Raiden, with a beta on mainnet in May 2020 for token transfers off chain ^[3]. Coil built Web Monetization on the Interledger Protocol, with a billion XRP behind it from Ripple, and ended its products in March 2023 ^[4].

Each of them proved the cryptography and struggled with the people. Lightning today has around 17,000 public nodes and 4,900 BTC of capacity ^[1], real, and small next to what it was meant to carry. Opening and closing a channel still needs a transaction on chain ^[1], every channel locks money up with one counterparty, and a person paying a coffee shop doesn't want to manage any of that. Agents don't mind. They'll open a channel, track a balance and sign a voucher every second without getting bored, which is why I think this time is different.

Who's built it for agents?

Three times, twice this year.

x402 batch settlement. Engineers from Coinbase and Cloudflare added it to x402 in May 2026 ^[5]. The agent commits funds to an escrow or channel, every HTTP request carries a cumulative signed "I owe you", the seller checks it with signature maths and no chain lookup, and value moves on chain "only when it is economically optimal", in bulk. Their own description is that it "brings the efficiency of payment channels directly into the x402 envelope" ^[5]. That's the design above, almost word for word.

MPP sessions. Stripe and Tempo's Machine Payments Protocol launched in March 2026 with sessions on Tempo, payment channels with signed vouchers off chain that bill per request or per token, and made them a formal part of the spec in June ^{[6] [7]}. There's a Solana version too, where the server "settles the highest accepted voucher on-chain whenever it chooses" ^[11]. It also shows what's hard about it. Within two weeks of MPP's launch, its reference library had a high-severity advisory titled "Tempo session close voucher bypass" ^[6]. Closing a channel is where the hard bugs are, in Lightning too.

L402. Lightning Labs' scheme is Lightning itself, a 402 response with a Lightning invoice and a macaroon, where the invoice's preimage is the proof you paid ^[8]. It's the oldest of the three (it started as LSAT), and its proposal to become a Lightning standard has been open for more than three years ^[8].

Why does almost nobody use them?

We list 26 tools with endpoints that take x402 payments, 15 fully and 11 in part ^[9]. One of them, BlockRun, is reported using batch settlement, live on Solana through PayAI Network since early October ^[10]. Most of the rest charge a transfer per call, and at least five of the listings sell prepaid credit or time instead, which is a channel where you trust the seller with the balance ^[9]. So the chain-clogging version is what's mostly live, and the channel versions are only starting to find sellers.

15 of the 18 tools that publish a price charge a cent or less for their cheapest call, and that's where paying on chain for every call makes the least sense ^[9].

I think the reasons are the same ones that held Lightning back. A channel with every tool locks the agent's money with every tool, and an agent might use thirty in a day. Someone has to watch for an old voucher being settled after a newer one. And the channels don't work across schemes. Servers can already advertise more than one, MPP accepts x402's simple payments and Lightning Labs' Aperture serves L402 and MPP side by side ^{[12] [8]}, but an x402 batch voucher isn't an MPP session voucher, so a seller still picks which channel its buyers need.

What would combining them look like?

Take the channel from Lightning, the stablecoin from the last few years so the bill doesn't move with the price of bitcoin, the HTTP envelope from x402 and MPP, and put a hub in the middle. The agent opens one channel, with the hub. The hub keeps a channel with each tool. Every call is a signed voucher on both sides, and each tool settles on chain when its bill is worth it. That's how Lightning routes payments between people who don't share a channel ^[1], and it's how card networks have worked for decades, with the settlement done in batches. The difference is trust. Lightning's routing nodes can't take what they forward, and a hub holding signed vouchers can, so the agent would be trusting the hub the way a shop trusts a card network.

That's what we want letme.dev to be for tools, one place an agent pays from, with the referral fee from the tool paying for the hub. Today it's a plan. letme picks the best tool for a job and says how to call it, and it doesn't take payments yet.

What I don't know is which of the three schemes wins, or whether it matters if a hub speaks all of them. If you run a tool that takes x402 and you've looked at batch settlement, I'd like to hear what stopped you.

References

Read on 16 October 2026.

1. [1]Wikipedia, Lightning Network. Source for the paper by Joseph Poon and Thaddeus Dryja in February 2015, the launch in 2018, the 17,000 public nodes and 4,900 BTC of capacity in March 2026, and opening and closing channels needing transactions on chain. en.wikipedia.org
2. [2]Bitcoin Wiki, Payment channels. Source for Satoshi Nakamoto's description of the technique in a personal email to a Bitcoin developer, quoted from Mike Hearn's bitcoin-dev post of 17 April 2013, Jeremy Spilman's design of 20 April 2013, and Mike Hearn and Matt Corallo's micropayment channels in bitcoinj, announced on 27 June 2013. en.bitcoin.it
3. [3]crypto.news, Raiden Network's Alderaan beta on Ethereum mainnet, 27 May 2020. Source for Raiden as Ethereum's counterpart to Lightning for token transfers, and the beta launch on 26 May 2020. crypto.news
4. [4]Jake Simmons, Bitcoinist, 3 February 2023. Source for Coil ending its products on 15 March 2023 and Ripple's grant of one billion XRP in August 2019. bitcoinist.com
5. [5]x402, Introducing x402 Batch Settlement, with Cam Whiteside (Cloudflare) and Carson Roscoe, Conner Swenberg, Josh Nickerson and Philippe d'Argent (Coinbase), 11 May 2026, updated 24 June 2026. Source for the escrow, the cumulative vouchers, verification without a chain lookup, settlement in bulk and the quotes. x402.org
6. [6]Anchor Terminal, the Machine Payments Protocol listing. Source for the launch on 18 March 2026, sessions becoming a formal intent on 17 June 2026 and the advisories in mppx later that month, including "Tempo session close voucher bypass". anchorterminal.com
7. [7]Anchor Terminal, the Tempo listing. Source for MPP sessions using payment channels and signed vouchers off chain, billed per request or per token. anchorterminal.com
8. [8]Anchor Terminal, the L402 listing. Source for L402's Lightning invoices and macaroons, the preimage as proof of payment, its former name LSAT, Aperture adding MPP alongside L402 in March 2026, and its proposal as a Lightning standard being open since June 2023. anchorterminal.com
9. [9]Anchor Terminal, x402 support across the directory, 16 October 2026. Source for the 26 tools that list x402 endpoints, 15 fully and 11 in part, the cheapest published price for each of the 18 that publish one, and the listings that sell prepaid credit, packs or time instead of charging per call (Apify, ScrapeGraphAI, Browserbase and Telnyx's two, with Browser Use and Hyperbrowser in part). anchorterminal.com
10. [10]Bitcoin.com News, Solana x402 lets AI agents batch USDC payments, 7 October 2026. Source for batch settlement going live on Solana through PayAI Network and BlockRun, and BlockRun using it in production. news.bitcoin.com
11. [11]MPP, Solana session payment method. Source for the channel deposit, the cumulative vouchers and the server settling the highest voucher whenever it chooses. mpp.dev
12. [12]MPP, EVM and x402 support, 8 June 2026. Source for MPP servers advertising both protocols and accepting x402 exact payments. mpp.dev