The Tool Desk
Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →AI agents pay for APIs, tools, data and web resources through a challenge-and-response transaction. The service first returns a machine-readable payment request; the agent checks its spending policy, signs a challenge-bound authorization, retries the request, and receives the resource only after verification and settlement succeed.
The quoted price is only one part of the cost. Network or gas charges, wallet and facilitator fees, and managed-service charges may apply. A failed settlement should block delivery, but a refund is not automatic: it depends on the merchant’s contract, capture timing, escrow or metering design, and dispute rules.
As an Amazon Associate I earn from qualifying purchases.
The payment loop an AI agent follows
Unlike a person entering card details, an autonomous agent needs a machine-readable way to discover the price, prove authorization and complete payment without an interactive checkout. The common sequence is:
- Request: The agent calls a paid API, tool, data feed or page.
- Challenge: The server returns HTTP
402 Payment Requiredor an equivalent response containing the amount, recipient, asset, network and payment method. - Policy check: The agent compares the request with rules such as an allowed merchant, maximum amount, permitted asset, session budget and expiry.
- Authorization: A wallet or payment component signs a credential tied to the challenge.
- Retry: The agent sends the signed payment payload in the protocol’s payment header or message object.
- Verification and settlement: The server or a facilitator checks the signature, amount, recipient, network, validity window and replay status, then settles the payment.
- Delivery: The paid response is released only when the required checks pass.
This ordering separates the seller’s quote from the agent’s authority to spend. A service can quote a price, but the agent’s local policy decides whether that price is acceptable. A valid signature proves authorization for the specified challenge; it does not give the agent unlimited permission to spend.
#1 Best Overall
What the challenge must establish
A useful challenge identifies the amount, payee, asset and network, and it gives the authorization a validity window or other anti-replay condition. If any of those values change between the quote and the signed response, the signature should no longer authorize the transaction. Implementations may add a request identifier, nonce or session identifier so that a captured payment message cannot be reused.
Why verification comes before delivery
Payment proof and resource access are separate decisions. A server may receive a malformed, expired or already-used authorization, or settlement may fail after the request is received. In those cases, the paid resource should remain withheld. Returning the resource before settlement would leave the seller exposed to unpaid requests.
How x402 turns HTTP 402 into pay-per-request commerce
x402 is the clearest example of an HTTP-native, pay-per-request flow. Its documentation describes a 402 response carrying payment requirements, a signed payment payload on retry, and independent or facilitator-based verification and settlement. Version 2 uses three named transport objects:
Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minute| Object | Purpose | When it appears |
|---|---|---|
PAYMENT-REQUIRED |
Communicates the seller’s requirements, including price, recipient, asset, network and supported method. | In the server’s payment challenge. |
PAYMENT-SIGNATURE |
Carries the agent’s signed authorization for the stated requirements. | On the agent’s payment retry. |
PAYMENT-RESPONSE |
Reports verification or settlement status and any resulting payment information. | After the server or facilitator processes the authorization. |
The specification is extensible rather than limited to one settlement pattern. It supports schemes such as exact, upto and batch settlement. An exact scheme charges the stated amount for the request. An upto design can authorize a ceiling and settle the measured amount, which is useful when usage is not known at the first request. Batch settlement can group multiple authorizations for later processing.
x402 leaves client-side budget management to the application. The protocol can describe what a seller requires, but the agent or its host must decide whether a call fits a per-request, per-session, per-agent or organization-wide limit. Builders should therefore treat a successful protocol exchange as authorization for one defined payment, not as permission to continue spending.
Rank #2
- Ideal for Gifting
- Ideal for a bookworm
- Compact for travelling
A typical x402 exchange
- The agent requests a protected endpoint.
- The endpoint replies with
402 Payment RequiredandPAYMENT-REQUIRED. - The agent validates the seller, network, asset, amount and expiry against its policy.
- Its wallet signs a
PAYMENT-SIGNATUREcontaining those challenge-bound terms. - The agent retries the same operation with the signature.
- The endpoint or facilitator verifies the signature and attempts settlement.
- The endpoint returns the resource with a
PAYMENT-RESPONSEor returns an error and withholds the paid content.
MPP, AP2 and x402 are related but not interchangeable
Several protocols use HTTP 402, but their message semantics and payment models differ. Comparing only the status code can hide important operational differences.
| Dimension | x402 | MPP | AP2 |
|---|---|---|---|
| Transport signal | Version 2 defines PAYMENT-REQUIRED, PAYMENT-SIGNATURE and PAYMENT-RESPONSE objects. |
Uses standard HTTP authentication semantics: WWW-Authenticate: Payment followed by Authorization: Payment. |
Focuses on signed mandates and accountable agent authorization rather than replacing x402 or MPP transport headers. |
| Payment model | Extensible schemes including exact, upto and batch settlement. | Documentation describes one-time charges and capped metered sessions. | Cryptographically signed mandates express what a user or principal authorized an agent to do. |
| Network and asset availability | Depends on the selected network, scheme and client implementation. | Depends on the merchant and payment implementation. | Depends on the payment rails used by the deployment. |
| Budget responsibility | Client or host application must enforce budgets. | Client or service must enforce caps and session limits. | Mandates can make authorization tamper-evident; deployments still need wallet and spending controls. |
MPP’s authentication-style headers can fit existing HTTP middleware, while x402’s objects make payment requirements and settlement responses explicit. Neither label alone tells you who settles, which assets are accepted, how retries are made safe, or whether refunds exist. Those details belong in the implementation and merchant contract.
How an agent authenticates a payment
Payment authentication is normally cryptographic rather than a password exchange. The signed credential is bound to the challenge’s amount, payee, asset, network and validity window. Verification checks that the signer controls the wallet or account, that the terms match the challenge, and that the authorization has not expired or been replayed.
Authentication versus authorization
- Authentication answers, “Which wallet or principal produced this signature?”
- Authorization answers, “Did that principal permit this amount, recipient, asset and time window?”
- Settlement answers, “Was the authorized transfer actually completed?”
A valid signature is not proof that funds were settled. Conversely, a settled transfer does not prove that the agent stayed within a user’s intended budget unless the deployment enforced that budget before signing.
Controls that keep an autonomous wallet bounded
Production deployments commonly combine wallet policy with identity, secret storage, per-agent budgets, session limits, expiry and optional human approval. A policy can restrict an agent to named merchants, approved networks or a maximum amount per call. A separate session cap can stop a loop from making thousands of individually small requests. Expiry limits the time during which an authorization can be presented.
AP2 approaches user authorization through cryptographically signed mandates. The goal is to make instructions tamper-evident and accountable: an agent can act within a mandate, while an auditor can inspect what was authorized. A mandate still needs a clear scope, spending ceiling, expiry and revocation path to be useful in practice.
Free tools Windows power users keep installed
One-click scans. No signup required.
Managed payment orchestration with AWS AgentCore Payments
A managed deployment can package protocol handling and wallet controls instead of requiring each agent developer to implement them. AWS AgentCore Payments can orchestrate x402 or MPP, enforce per-session spending limits and expiry, retrieve payment tokens through GetResourcePaymentToken, and integrate Coinbase CDP or Stripe (Privy) embedded wallets.
This layer can centralize identity, token retrieval and policy enforcement, but it does not make every payment free or every merchant refundable. AWS documentation states that a failed signing operation after budget deduction does not deduct the failed payment. That protects the failed attempt; it does not reverse a payment that was successfully signed and settled before a later application error.
What an AI agent actually pays
The amount in the challenge is the seller’s price. The final debit can include several additional components:
| Cost component | What it covers | Who may charge it |
|---|---|---|
| Resource price | The API call, tool execution, data item or page requested. | The merchant or resource operator. |
| Network or gas fee | Transaction processing on the payment network. | The network, usually paid from the wallet or arranged through a sponsoring service. |
| Wallet fee | Wallet custody, signing or transaction handling where applicable. | The wallet provider. |
| Facilitator fee | Verification, submission or settlement services performed for the merchant or agent. | The facilitator, if its terms include a charge. |
| Cloud or orchestration charge | Managed token retrieval, policy enforcement or runtime infrastructure. | The cloud or managed-service provider. |
The x402 website describes the protocol as free for customer and merchant while saying users pay nominal payment-network fees. That is a protocol-fee statement, not a guarantee that every network, wallet, facilitator or managed service has no charge. Read the deployment’s fee schedule and determine whether the agent, merchant or a facilitator supplies the network fee.
Rank #4
Why tiny charges are technically significant
AWS explains that traditional payment systems often impose minimum fees that make transactions below $0.01 uneconomic. Stablecoins and HTTP-native payment protocols target that gap by allowing machine-to-machine settlement without a conventional card checkout. The economic result still depends on the network’s fee, confirmation requirements, exchange-rate exposure and the cost of operating the wallet.
An IETF Internet-Draft for an AEPP design proposes a micro-fee not exceeding 0.4% per execution. That is a draft proposal, not a universal rule for x402, MPP or AP2. There is no authoritative universal industry average fee published for AI-agent micropayments.
Interpreting reported activity
The x402 site displayed 75.41 million transactions and $24.24 million in volume on October 15, 2026. Those are volatile, platform-reported dashboard snapshots, not an independently audited measure of the entire market or a guarantee of future usage.
Who pays gas and network fees?
In a self-custodied setup, the wallet controlled by the agent or its operator normally supplies the network fee. A facilitator can submit or settle transactions on behalf of the merchant, and a managed service can arrange token retrieval or fee handling, but someone must fund the underlying network transaction. The challenge amount may therefore be lower than the total wallet debit.
Recommended Free Tools
Before enabling autonomous spending, establish whether the quoted amount is inclusive of network costs, whether the network fee is estimated or variable, which asset pays it, and what happens when the wallet lacks that asset. A policy that caps only the seller’s price can still permit unexpected fee spending unless it also limits total transaction cost.
Best Value
- It can be a gift option
- Comes with secure packaging
- Helpful in various ways
What happens when payment fails?
A payment can fail before signing, during verification, during network settlement or after the tool has started. Each stage needs a different response.
| Failure stage | Expected result | Agent action |
|---|---|---|
| Policy rejection | No authorization is produced and no payment should be attempted. | Stop, log the reason and request approval or a different option. |
| Signing failure | No valid payment credential exists. | Retry only after checking wallet availability, limits and expiry. |
| Verification failure | The server rejects an invalid, mismatched, expired or replayed credential. | Do not resend the same signature; obtain a fresh challenge. |
| Settlement failure | The paid resource remains withheld. | Use the protocol’s error response, check whether any transfer occurred, and avoid duplicate payment until status is known. |
| Tool or application failure after settlement | Settlement may have succeeded even though the resource was not usable. | Follow the merchant’s refund or dispute policy; protocol-level failure handling does not guarantee reimbursement. |
The x402 MCP transport specifies an error result when settlement fails, including a case where a tool has already run; the tool content is then withheld. This is delivery protection, not a consumer-refund guarantee.
Refunds are an application-level promise
Core x402 defines payment requirements, authorization, verification and settlement. It does not define one universal refund policy, and the same limitation applies when a service combines a payment protocol with its own application logic. A refund depends on the merchant contract and the settlement design.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Questions that determine refund eligibility
- Were funds captured before execution, after execution or in stages?
- Was the authorization exact, capped, metered or held in escrow?
- Did the tool return a usable result, a partial result or an error?
- Who can authorize a full or partial refund?
- How long does the customer have to open a dispute?
- Was enough liquidity reserved to send the refund?
An upto or metered scheme can reduce overpayment by settling measured usage, but it does not automatically refund an unusable result. Escrow-oriented designs can delay final fulfilment and support partial refunds. Agent Commerce Kit lists refund-management patterns, and a 2026 escrow-oriented research proposal describes delayed fulfilment and partial refunds; these are higher-level patterns, not evidence that every x402 or MPP merchant supports automatic reimbursement.
What the agent should record
For each paid call, retain the challenge, signed authorization, policy decision, settlement identifier, delivered-resource status and any error response. These records let an operator distinguish “never paid,” “paid but not delivered” and “delivered but defective.” Without that separation, an automatic retry can create a second charge while the first transaction is still pending.
Implementation checklist for safer agent payments
- Define the budget: Set per-call, per-session, per-agent and organization limits, including network fees.
- Allow-list counterparties: Restrict merchants, networks and assets where possible.
- Require fresh challenges: Reject stale or reused payment requirements and signatures.
- Set expiry: Give authorizations short, explicit validity windows appropriate to the tool.
- Choose payment semantics: Use exact payment for known prices, capped metering for variable usage, or batching when settlement overhead justifies it.
- Separate keys: Use an agent-specific wallet or delegated credential instead of an unrestricted personal wallet.
- Add approval thresholds: Route unusual amounts, new merchants or new networks to a person.
- Make retries idempotent: Persist request and settlement identifiers so a timeout does not trigger duplicate payment.
- Publish refund terms: State capture timing, eligibility, partial-refund rules, dispute windows and expected processing time.
- Monitor settlement: Alert on repeated failures, budget exhaustion, fee spikes and calls that settle without a usable response.
How to compare an agent-payment option
Evaluate the implementation, not just its protocol name. The most consequential questions are:
- Message semantics: Which headers or objects carry requirements, authorization and settlement status?
- Payment model: Is the charge exact, capped, metered or batched?
- Assets and networks: Which currencies and networks are supported, and who pays their fees?
- Verification and settlement: Is the merchant self-verifying, or does a facilitator submit and settle?
- Budget and expiry: Are limits enforced locally, by a managed service, or not at all?
- Replay protection: What binds a signature to one request and prevents reuse?
- Privacy and identity: What wallet, user or agent information is disclosed to the merchant and facilitator?
- Refund governance: Is there escrow, a documented dispute process, or only a discretionary merchant policy?
- Standards maturity: Is the feature a deployed specification, an extension, or a draft proposal?
Bottom line
AI-agent micropayments are an automated HTTP transaction loop: challenge, policy check, signed authorization, retry, verification, settlement and delivery. x402 makes that loop explicit with payment objects and extensible settlement schemes; MPP uses payment-aware HTTP authentication; AP2 adds signed mandates for accountable user authorization. Authentication proves that a bounded wallet or principal approved specific terms, while settlement proves that money moved.
The seller’s quoted amount is not necessarily the total cost, because network, wallet, facilitator and cloud charges may remain. Failed settlement should prevent access to the paid resource. A refund, however, requires an application-level promise backed by capture rules, escrow or metering, dispute procedures and available liquidity. Anyone deploying an autonomous payer should therefore cap total spend, bind every signature to a fresh challenge, make retries idempotent and read the refund contract before allowing the agent to act unattended.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
