Skip to main content

Payments for AI agents: Atto, Nano, USDC and Lightning compared

· 15 min read
Felipe Rotilho
Software Engineer @ Atto

An AI agent needs one data lookup. The price is $0.001. Before the provider can return a result, the agent needs a way to pay that leaves the sale worth making.

We build Atto for small, direct agent payments: no protocol transfer fees, first-party MCP, CLI and n8n integrations, and public node access. Its wallet tools are available today; a complete paid-HTTP service still needs custom payment logic.

USDC brings dollar pricing and existing x402 support. Lightning serves bitcoin payments through L402. Nano offers another feeless option with community agent tools. For AI agent micropayments, the useful choice depends on the service, the payment size and the integration you want to run.

Capabilities, pricing and source snapshots checked on November 2, 2026. Individual timing measurements have the dates shown below.

Compare payment networks for AI agents​

Start with what the service accepts and what each request costs.

Your priorityConsiderCheck before starting
Very small direct transfers without a protocol transfer feeAtto or NanoAtto's CLI/MCP tools are in beta; accepting paid API calls needs custom service logic.
Dollar pricing with an existing paid-HTTP flowUSDC on Base or Solana through x402Compare the facilitator's tariff with the price of each call.
Bitcoin payments for API accessLightning through L402Choose a wallet backend and check routing and provider fees.

The agent-payment guide explains how a wallet payment fits into an API purchase and walks through a first bounded Atto payment.

The fee comparison covers $0.001 calls paid individually. USDC scores use Coinbase's CDP Facilitator after its monthly free allowance; Atto and Nano use direct transfers, and Lightning uses funded channels.

Timings come from different network measurements and do not measure complete API purchases. Base uses L2 preconfirmation; Lightning uses off-chain settlement. For Atto and Nano, a confirmed send is enough to deliver the purchase; receiving it later makes the merchant's funds spendable.

Scroll horizontally to see all criteria.

Payment options for AI agents
Payment optionTotal scoreTiny-transfer economicsPrice vs ATHAgent integrationLayer confirmation
AttoDirect ATTO transfers4.2 out of 5.5 out of 5.
No protocol transfer fee
1.7 out of 5.
82.1% below ATH
5 out of 5.
Tools
First-party MCP, CLI and n8n. CLI/MCP beta.
Backend
Public Node and Worker defaults.
Paid API
Custom service logic; x402 planned.
158 ms5 out of 5.
Node confirmation average. Nov 1, 2026.
NanoDirect XNO transfers3.7 out of 5.5 out of 5.
No protocol transfer fee
1 out of 5.
99.1% below ATH
4 out of 5.
Tools
Community CLI/MCP.
Backend
Public RPC options; no own node required.
Paid API
Community x402; check confirmation.
331 ms4.6 out of 5.
Probe confirmation average. Nov 1, 2026.
USDC on BaseCDP exact, paid tier3.7 out of 5.1 out of 5.
$0.001 fee on a $0.001 sale
5 out of 5.
Stablecoin
4.25 out of 5.
Tools
AgentKit; x402 SDKs.
Backend
Hosted CDP; account policies configured.
Paid API
Available through x402 exact.
290 ms4.7 out of 5.
L2 preconfirmation average. Aug 12, 2026.
USDC on SolanaCDP exact, paid tier3.7 out of 5.1 out of 5.
Same CDP per-settlement fee
5 out of 5.
Stablecoin
4.25 out of 5.
Tools
AgentKit; x402 SDKs.
Backend
Hosted CDP; account policies configured.
Paid API
Available through x402 exact.
399 ms4.4 out of 5.
Rooted-finality average. Nov 2, 2026.
Bitcoin LightningFunded payment channels2.8 out of 5.3 out of 5.
Routing and provider fees vary
3.5 out of 5.
38.1% below ATH
3.5 out of 5.
Backend
Self-operated wallet backend.
Paid API
L402 and paid MCP available.
1.86 s1 out of 5.
Provider-reported settlement average. 2026 pilot.

Ratings and timing methods explained below.

Total score averages the four displayed ratings equally. Agent integration scores are provisional, with setup and maintenance carrying 60% of the weight. They assess the named tools and backends. Price vs ATH measures position relative to the dollar all-time high, not dollar stability. Prices are a November 2 snapshot; stablecoins receive 5 by convention.

When a $1 workload leaves nothing to run the service​

1,000 calls at $0.001 each earn $1. After CDP's monthly free allowance, one successful exact settlement per call costs $0.001, adding up to $1 in fees and $0 left before serving the requests. The first 1,000 monthly transactions are free, verification is free, and the fee already covers settlement gas. CDP pricing, pricing FAQ.

Under that paid tariff, individually settled calls at or below $0.001 cannot cover service costs. Larger mini payments have more room: the same fee takes 10% of a $0.01 sale or 1% of a $0.10 sale. Batching or another facilitator changes the economics.

Atto and Nano charge no protocol transfer fee. Funding, conversion, computational work and infrastructure still cost resources. USDC makes dollar pricing easier, while adding issuer and redemption conditions. USDC terms.

What each option offers​

Atto: feeless payments with first-party agent tools​

Atto suits individual payments where a protocol fee would consume too much of the sale. The first-party MCP, CLI and n8n tools use public services, so developers can start without finding or hosting a node.

The CLI/MCP wallet starts read-only and lets you approve per-payment and rolling spending limits locally. Those packages are in beta. A complete paid-API flow still needs custom quoting, payment matching and delivery logic; the x402-style flow and demonstration endpoint remain planned work. ATTO has a floating dollar value.

Nano: feeless payments with community integrations​

Nano also fits small, direct payments. Its community Feeless402 CLI/MCP provides public RPC defaults, and Nano's agent page links x402 tooling. Running your own node is optional.

Check confirmation handling before delivering a purchase: the linked x402 implementation can return a block hash before confirmation. Its implementation notes explain deployment limits. XNO also has a floating dollar value.

USDC on Base: dollar pricing and existing x402 support​

USDC on Base fits services that price in dollars and already support x402. The client SDK and hosted facilitator provide a paid-HTTP path without operating a Base node.

The CDP fee example above makes this setup better suited to larger small payments once the free allowance is exhausted. The table accepts an L2 preconfirmed receipt for delivery; a seller requiring stronger finality must wait longer.

USDC on Solana: dollar payments for Solana services​

USDC on Solana serves the same dollar-pricing need through another network. It is useful when the seller and agent already use compatible Solana x402 tooling.

Configure the wallet and recipient token accounts for that path. Native transfers have network fees; a hosted facilitator can cover gas through its own tariff. The CDP per-settlement fee applies here too, so a fast network alone does not solve the $0.001 sale problem.

Bitcoin Lightning: paid access through a Lightning wallet​

Lightning fits bitcoin services using L402. Lightning Labs' lnget CLI/MCP can pay a challenge and retry the request; Alby MCP connects agents and n8n workflows to a compatible wallet.

A wallet provider can remove the need to run your own node. Self-hosting adds channel and liquidity management; routing and provider fees vary. Bitcoin's dollar value moves with the market.

Start with the purchase​

Choose a payment option the service accepts, then test a small purchase and its timeout recovery. A prepaid service balance may also be simpler when calls can be metered internally.

To try Atto, follow the agent-payment guide for a first bounded payment. The integration guides help you choose MCP, CLI or n8n; the faucet can fund a small wallet. A paid service must connect payment confirmation to its own purchase and delivery flow.

Sources and methodology​

Rating rules, measurement sources and implementation notes

Total score. The second column is the equal-weight average of tiny-transfer economics, price vs ATH, agent integration and layer confirmation. It uses the displayed component ratings, including the confirmation rating rounded to one decimal, then rounds the average to one decimal. All four ratings are required; an unscored component leaves the total unscored.

Comparison scope. The economics scores compare individual $0.001 payments: direct Atto/Nano transfers, CDP exact settlements after the monthly free allowance on Base/Solana, and Lightning payments through funded channels. Economics scores are 5 for no protocol transfer fee, 3 for route-dependent costs and 1 when the stated gateway fee consumes the sale. They are not total operating-cost estimates or production-reliability ratings.

Agent integration. These provisional editorial scores assess how easy a named agent-wallet integration is to get working and maintain. The starting point is a funded wallet: obtaining funds is excluded, while configuring the integration and operating its required backend are included. This is a documentation and code assessment, not a runtime usability test or a production-reliability rating. Seller paid-API readiness remains a separate field.

Setup and maintenance each contribute 1.5 points; the ready agent interface and externally enforced spending controls each contribute 0.5. Clear payment outcomes and safe recovery each contribute another 0.5, combined in the last criterion below. Each individual criterion—including outcomes and recovery separately—receives full, half or no credit against its maximum. The setup and maintenance judgments are provisional.

Evaluated integration pathSetup /1.5Maintenance /1.5Agent interface /0.5Spending controls /0.5Results & recovery /1Total /5
ATTO — Atto MCP1.51.50.50.515
XNO — Feeless402 MCP1.51.50.500.54
USDC/Base — AgentKit1.51.50.50.50.254.25
USDC/Solana — AgentKit1.51.50.50.50.254.25
BTC/Lightning — Alby MCP + Hub0.750.750.50.513.5
  • Atto MCP: public backend defaults, operator-approved spending caps, structured payment states and durable request-ID reconciliation. The reviewed spending checks and payment workflow support the capability scores. CLI/MCP are beta.
  • Feeless402 MCP: public RPC defaults keep setup and backend operation light. Its tool price cap is an agent-controlled argument, not an owner-enforced budget. The selected payment flow has partial confirmation reporting and recovery coverage.
  • AgentKit on Base and Solana: these scores assume separately configured CDP account policies, supported by the EVM and Solana policy engines. The reviewed ERC20 and SPL transfer tools return identifiers but need application work for complete outcome handling and uncertain retries. The policies establish permissions or per-transaction limits, not cumulative daily budgets.
  • Alby MCP + Hub: wallet budgets, payment lookup and same-invoice duplicate protection earn capability credit. The assessed Hub setup is self-operated, so backend setup and maintenance receive partial credit. Payment handling and recovery tests support the results/recovery score. This does not grade every Lightning wallet or provider.

Layer-confirmation bars. The shortest listed duration receives 5 and the longest receives 1. Intermediate values use 5 − 4 × (duration − shortest) ÷ (longest − shortest), rounded to one decimal. If all listed durations are identical, each receives 5. The resulting indices are Atto 5, Nano 4.6, Base 4.7, Solana 4.4 and Lightning 1. Different timing boundaries and assurances mean these are source-specific observations rather than a controlled network ranking.

Price vs ATH. Native-asset scores use 1 + 4 × (current USD price ÷ recorded USD all-time high), capped at 5 and rounded to one decimal. Stablecoins receive 5 by convention. This measures price position, not dollar stability, volatility or safety.

The CoinGecko API snapshot was fetched on November 2, 2026 at 09:59 UTC:

  • ATTO: $0.00003932 versus a $0.00021937 ATH; provider update 02:50 UTC that day.
  • XNO: $0.319043 versus a $33.69 ATH; provider update 09:58 UTC.
  • BTC: $78,023 versus a $126,080 ATH; provider update 09:58 UTC.

ATTO's provider observation was about seven hours old at retrieval; the other two were about a minute old. These are dated inputs, not automatically updating quotes.

Timing sources. The table uses observed mainnet activity, with these boundaries:

  • Atto: 158.34 ms daily mean for November 1, from node arrival to confirmed storage across two node databases and all account-update types. This excludes earlier wallet/client submission and is not Send-only. Metric source, measurement scope.
  • Nano: 331 ms from NanoSpeed's latest daily point dated November 1 and its own single-block probes. Probe methodology.
  • Base: 290.3 ms mean over 50 successful mainnet USDC self-transfers in US-east on August 12, 2026, timed from submission to Flashblock receipt. This is historical L2 preconfirmation, not full sealed-L2 inclusion, irrevocable finality or a complete CDP request. The published aggregate has no individual timings or measurement harness. Immutable FillBench result, Base receipt semantics.
  • Solana: 399.254 ms mean across 402,618 observations in OpenChainBench's 24-hour window as of November 2 at 10:09 UTC. The collector times first slot visibility until a root notification covers it. This excludes earlier submission and CDP processing. Results, collector.
  • Lightning: Voltage reports a 1.86-second mean for 237,000 payments over a 30-day production pilot, published in 2026. The traffic was predominantly Cash App; exact timer events and raw records are undisclosed. This is provider-reported off-chain settlement rather than L402 API delivery. Report, pilot announcement.

Download dated observation data for the bar inputs and source references. It retains an additional, unscored Base L1 observation: one existing USDC transfer took 20 minutes 38 seconds from first observed L2 inclusion to Ethereum-backed batch finality on November 2. Polling timestamps, successful receipt and unchanged block-hash checks are in that download. This stronger assurance is excluded from the L2 score.

The Atto, Nano and RSNano benchmark repository contains reproduction commands and raw data for controlled single-host, one-voter tests. Those tests are separate from the mainnet observations above.

Implementation notes. A wallet tool, a payment network and a paid-HTTP protocol handle different steps. x402 requests payment through HTTP 402; L402 uses a Lightning invoice and payment proof. For Atto/Nano, match the confirmed send's recipient, amount and purchase before delivery; Receive only makes funds spendable. Check existing payment status after a timeout before starting another payment.

Atto's CLI/MCP share public service defaults. Their limits constrain that wallet profile, not arbitrary programs with unrestricted computer access. n8n signing credentials do not inherit those limits. Commons supports custom applications. Nano's x402nano helper needs an RPC URL; its production integration guide covers key management rather than relying on the internal node wallet.

Direct Base transfers have execution and publication fees; CDP includes settlement gas in its own charge. CDP batching can change per-call economics, but its EVM batch scheme needs a separate supported integration because createX402Server does not register it. Current support.

Solana's fee payer can differ from the sender. A new token account requires a refundable storage deposit; on closure, the owner or close authority selects its destination. Lightning self-hosting requires sufficient outbound/inbound liquidity; Aperture supplies the L402 proxy path.