Mache deine bestehende HTTP API für AI Agents verfügbar, damit sie diese finden, aufrufen und bezahlen können – ohne dein Backend ändern zu müssen.
Auch bekannt als devlab-group/agent-commerce, @devlab.group/agent-commerce
- Stars
- 71
- Tools
- —
- Downloads, 30 Tage
- 1.614
- Neueste Version
- —
- Letzter Commit
- 10. Nov. 2026
- Lizenz
- Apache-2.0
Was ist Agent Commerce Gateway MCP?
Agent Commerce Gateway ist ein MCP server. Mache deine bestehende HTTP API für AI Agents verfügbar, damit sie diese finden, aufrufen und bezahlen können – ohne dein Backend ändern zu müssen.
Zusammenfassung generiert aus dem README des Projekts
Über Agent Commerce Gateway MCP server
Das Agent Commerce Gateway übernimmt die Aufgabe, eine bestehende HTTP API für AI Agents zugänglich zu machen, damit diese sie finden, aufrufen und bezahlen können, ohne dass das Backend geändert werden muss. Entwickler lassen es in ihrer eigenen Infrastruktur laufen und beschreiben jeden Endpoint in YAML oder generieren Entwürfe für Resource-Definitionen über den OpenAPI-Import des CLI. Das Node >= 22 Package liefert eine agent-commerce Command Line mit init, validate, start, doctor, reconcile, resolve, demo, import openapi und version sowie eine Library, um das Gateway in deinen eigenen Prozess einzubetten.
Resources können über einfaches HTTP, MCP oder das experimentelle A2A bereitgestellt werden, und dedizierte Resources können experimentelle ACP-Checkout-Operationen implementieren. Bezahlte Angebote werden über x402 oder das experimentelle MPP abgewickelt und können zusätzlich ein AP2-Mandat erfordern, das zeigt, dass der Käufer den Kauf bestätigt hat. Zahlungen gehen direkt an das Wallet des Merchants; das Gateway hält weder den Private Key des Käufers noch den des Merchants. Protocol-, Payment- und Authorization-Peers sind optionale Subpath-Imports, sodass ein kostenloses HTTP-Deployment schlank bleibt, während das Importieren eines Subpaths ohne seinen Peer beim Laden fehlschlägt und das fehlende Package benennt. Der Quickstart benötigt Node, npm und Docker, aber keine API Keys oder echtes Geld, da er einen temporären Stack aus Anvil, MockUSDC, einer Demo-Merchant-API, dem Gateway und einem Dashboard bereitstellt.
Jeder Request durchläuft eine Execution-Pipeline, die Authorization, Payment, einen begrenzten Backend-Call und einen SQLite-Receipt-Store abdeckt, sodass die Durchsetzung nicht vom Adapter abhängt. Commands akzeptieren eine env-file vor dem Lesen der Config, und ein reconcile- sowie ein doctor-Command helfen dabei, das Setup zu verifizieren.
Was du damit machen kannst
- 1Endpoints als MCP Tools bereitstellenLass Claude einen bestehenden API-Endpoint als MCP Tool entdecken und aufrufen, nachdem du ihn in YAML beschrieben hast.
- 2Agenten pro API-Call abrechnenErlaube einem Agenten, einen kleinen Stablecoin-Betrag für einen Premium-Report zu bezahlen und diesen erst nach dem Settlement zu erhalten.
- 3Manuelle Resource-Config überspringenGeneriere Entwürfe für Resource-Definitionen aus einem OpenAPI-Dokument, anstatt jeden Endpoint von Hand zu schreiben.
- 4Stack vor dem Livegang prüfenFühre den doctor-Command gegen eine Demo-Config aus, um sicherzustellen, dass der gesamte Stack funktioniert, bevor du die Produktion anfasst.
- 5Aufgezeichnete Payment-Receipts prüfenUntersuche den Receipt-Store und gleiche settled Payments mit den erwarteten On-Chain-Balance-Änderungen ab.
Geschrieben aus dem README des Projekts. Jeder Use Case wurde gegen ein Zitat daraus geprüft.
So installierst du Agent Commerce Gateway MCP
Formate geprüft am 10. Nov. 2026Zu Cursor hinzufügenZu Goose hinzufügenAgent Commerce Gateway MCP in Claude Code installieren
$ claude mcp add agent-commerce -- npx -y @devlab.group/agent-commerceAgent Commerce Gateway MCP in Cursor installieren
~/.cursor/mcp.json
{
"mcpServers": {
"agent-commerce": {
"command": "npx",
"args": [
"-y",
"@devlab.group/agent-commerce"
]
}
}
}Agent Commerce Gateway MCP in VS Code installieren
.vscode/mcp.json
{
"servers": {
"agent-commerce": {
"type": "stdio",
"command": "npx",
"args": [
"-y",
"@devlab.group/agent-commerce"
]
}
}
}Agent Commerce Gateway MCP in VS Code CLI installieren
$ code --add-mcp '{"name":"agent-commerce","type":"stdio","command":"npx","args":["-y","@devlab.group/agent-commerce"]}'Agent Commerce Gateway MCP in Claude Desktop installieren
claude_desktop_config.json
{
"mcpServers": {
"agent-commerce": {
"command": "npx",
"args": [
"-y",
"@devlab.group/agent-commerce"
]
}
}
}Agent Commerce Gateway MCP in Codex installieren
$ codex mcp add agent-commerce -- npx -y @devlab.group/agent-commerceAgent Commerce Gateway MCP in Codex config.toml installieren
~/.codex/config.toml
[mcp_servers.agent-commerce] command = "npx" args = ["-y", "@devlab.group/agent-commerce"]
Agent Commerce Gateway MCP in Gemini CLI installieren
$ gemini mcp add -s user agent-commerce npx -- -y @devlab.group/agent-commerceAgent Commerce Gateway MCP in Gemini settings.json installieren
~/.gemini/settings.json
{
"mcpServers": {
"agent-commerce": {
"command": "npx",
"args": [
"-y",
"@devlab.group/agent-commerce"
]
}
}
}Konfiguration
DISABLE_TELEMETRY
| Name | Beschreibung | Typ |
|---|---|---|
| DISABLE_TELEMETRY | — |
Automatisierte Checks
Geluften am 13. Nov. 2026Vollständiger Bericht →Unsere automatisierten Checks bestanden. Das ist kein Security-Audit: Lies den Code und gib dem Server nur den minimal notwendigen Zugriff. Wie die Checks funktionieren
- Source Repository verknüpft und erreichbarBestanden
- Open Source Lizenz (OSI)Bestanden
- Commit oder Release im letzten JahrBestanden
- Package-Name ähnelt keinem populären PackageBestanden
- Remote Endpoint nutzt HTTPSNicht geprüft
- Destruktive Tools sind markiertNicht geprüft
- Keine versteckten Anweisungen an das ModellBestanden
- Keine kritischen Advisories im Paket oder seinen AbhängigkeitenNicht geprüft
README
Auf GitHub ansehenWhat it is, in ten seconds
You already have an HTTP API. Agent Commerce Gateway lets AI agents discover, call and pay for it without adding those protocols to your backend.
Run the gateway in your infrastructure and describe each endpoint in YAML, or generate draft resource definitions from OpenAPI. Resources can be exposed over HTTP, MCP or experimental A2A; dedicated resources can implement experimental ACP checkout operations. Paid resources use x402 or experimental MPP and can also require an AP2 mandate proving that the buyer approved the purchase. Funds go directly to the merchant wallet, and the gateway needs neither the buyer's nor the merchant's private key.
Your existing API → Agent Commerce Gateway → AI Agent
MCP · A2A · ACP · x402 · MPP · AP2 · receipts · doctor
Demo
[agent] Agent Commerce Gateway: deterministic buyer demo
[buyer] balance before 100 MockUSDC
[agent] merchant balance before 0 MockUSDC
[gateway] connected to http://localhost:8080/mcp
[agent] discovered 2 tool(s): weather_basic, market_report
[agent] paid tool: "market_report", free tool: "weather_basic"
[gateway] 402 payment required: 0.01 USDC to 0x7099…79C8
[buyer] challenge checked before signing: pays 10000 units of 0x5fbd…0aa3 to 0x7099…79C8
[buyer] signed x402 payment proof (LOCAL DEVELOPMENT ONLY - DO NOT FUND)
[gateway] delivered "market_report"
[receipt] id=receipt_484b…1eba requestId=mcp_4292…c930 status=settled settlementTx=0x0b19…1c7f
[buyer] balance after 99.99 MockUSDC (Δ -0.01)
[agent] merchant balance after 0.01 MockUSDC (Δ 0.01)
[agent] on-chain balance changes match the expected payment amount
[gateway] delivered free tool "weather_basic": {"city":"Paris","temperatureC":-4,…}
[agent] demo complete: free and paid delivery both verified
This excerpt of npm run demo:agent omits the address lines and shortens long
values. Balances, IDs and hashes differ from run to run.
The dashboard at http://localhost:5173 polls the gateway's events route and shows the same request within a few seconds.
Install
npx @devlab.group/agent-commerce --help # no install needed
npm install -g @devlab.group/agent-commerce # or install the `agent-commerce` binary
agent-commerce doctor
Requires Node >= 22. The package includes the agent-commerce CLI
(init, validate, start, doctor, reconcile, resolve, demo,
import openapi, version) and a library for embedding the gateway.
The default install excludes optional protocol, payment and authorization
peers. The OpenAPI parser is a regular dependency because import is part of the
CLI.
agent-commerce start --config config.yaml --env-file .env
start runs the configured gateway in the foreground until SIGTERM or SIGINT.
Install the optional peers for the features you enable, listed below. Missing
startup imports stop the gateway with an error naming the package; missing
Solana, NEAR or CDP peers fail provider health checks or payment requests.
start, validate, doctor, reconcile and resolve accept --env-file <path> to
load environment variables before reading the config. Variables already set in
the environment take precedence. To embed the gateway in your own process, use
the library:
import { createGateway, loadConfig, receipts } from '@devlab.group/agent-commerce';
const config = await loadConfig({ path: 'config.yaml' });
const store = receipts({ path: './receipts.sqlite' });
// Initialize before use so an incompatible SQLite schema fails at startup
await store.init();
const gateway = await createGateway({
config,
store,
paymentProviders: [],
protocolAdapters: [],
});
const { url } = await gateway.listen();
Optional peers
Components with extra dependencies use separate subpaths, so a gateway serving free HTTP resources does not install MCP, EVM, JOSE or SD-JWT packages.
| You want | Install | Import |
|---|---|---|
| gateway, config, receipts, A2A/ACP, CLI | @devlab.group/agent-commerce |
from '@devlab.group/agent-commerce' |
| expose resources as MCP tools | + @modelcontextprotocol/server |
from '@devlab.group/agent-commerce/mcp' |
| accept x402 payments | + @x402/core @x402/evm viem |
from '@devlab.group/agent-commerce/x402' |
| accept x402 on Solana | x402 peers + @x402/svm @solana/kit |
(no import - loaded for a Solana offer) |
| accept x402 on NEAR | x402 peers + @x402/near |
(no import - loaded for a NEAR offer) |
| accept MPP payments | + mppx viem @x402/core @x402/evm |
from '@devlab.group/agent-commerce/mpp' |
| verify AP2 mandates, sign checkout JWTs | + jose @sd-jwt/core canonicalize |
from '@devlab.group/agent-commerce/ap2' |
The x402 entry needs @x402/core, @x402/evm and viem even when every
offer is on Solana or NEAR. MPP settles through an x402 facilitator, so its
entry also needs the x402 peers.
npm install @devlab.group/agent-commerce @modelcontextprotocol/server @x402/core @x402/evm viem
import { mcp } from '@devlab.group/agent-commerce/mcp';
import { x402 } from '@devlab.group/agent-commerce/x402';
import { ap2 } from '@devlab.group/agent-commerce/ap2';
AP2 peers remain optional because a deployment without mandate-gated resources does not need JOSE or SD-JWT support.
Peers are pinned exactly: x402's schemas and EIP-712 domains cross this boundary, so a version skew is a correctness problem rather than a convenience one. Importing a subpath without its peer fails at load time and names the missing package.
Testing a custom store
createGateway accepts a custom ReceiptStore. The experimental ./testing
entry exports conformance cases to test it. The cases use node:assert/strict
and import no test runner or optional peers, so you can wrap them with your
runner:
import { receiptStoreConformanceCases } from '@devlab.group/agent-commerce/testing';
import { it } from 'vitest';
const cases = await receiptStoreConformanceCases(({ clock }) => ({
store: createMyStore({ clock }),
}));
for (const c of cases) it.skipIf(c.skip !== undefined)(c.name, () => c.run());
Have the factory return a new, empty store for each case. The suite calls
init() and close(). Pass the supplied clock to the store so expiry cases
can advance time without waiting.
Each optional method group, such as authorization holds, must be implemented
in full or omitted; one case checks this. Cases that need missing methods are
marked to skip. Durability cases also skip unless the factory returns reopen(),
which must close the current store and open another over the same data.
Concurrency cases submit competing reservations, resolutions, idempotency
claims and pending-task takes, then check that exactly one caller wins.
Integrate with an AI coding agent
The repository includes an experimental Agent Skills
skill, agent-commerce, for adding the
gateway to an existing API. It covers Base Sepolia configuration, startup and
checks through an unpaid 402 challenge. Its instructions keep secrets out of
chat and leave payment signing to the user. The skill has automated CLI and
config checks.
# Claude Code
/plugin marketplace add devlab-group/agent-commerce
/plugin install agent-commerce-gateway@agent-commerce
# Codex, Cursor and other agents (a third-party installer)
DISABLE_TELEMETRY=1 npx skills add devlab-group/agent-commerce --skill agent-commerce
Quickstart
Requirements: Node >= 22, npm 10, Docker. Nothing else - no API keys, no real money, no manual blockchain setup.
git clone <repo> && cd agent-commerce
npm install
docker compose up
Then, in a second terminal:
npm run agent-commerce -- doctor --config config-demo.yaml # verify the whole stack
npm run demo:agent # watch an agent buy something
Linux: if your UID/GID is not 1000, export DOCKER_UID=$(id -u) DOCKER_GID=$(id -g) before docker compose up. This keeps the generated deployment manifest host-writable. Docker Desktop does not need this.
The disposable local stack includes Anvil, MockUSDC, a demo merchant API, the gateway and a dashboard.
To stop and wipe state: docker compose down -v.
agent-commerce demo
From a repository clone, one command starts the stack and runs the buyer agent:
npm run agent-commerce -- demo
It starts the stack with docker compose up --build --detach --wait, which
returns once every service with a health check is healthy, then runs
npm run demo:agent. A failing step stops the run with exit code 1, after
printing the end of its error output and the command to rerun it by hand.
demo does not run doctor. The Linux note above applies, and the stack
keeps running in the background until docker compose down -v.
How it works
┌──────────────────────────────────────────────────────────────────────────────┐
│ AI Agent │
└───────────────────────────────────────┬──────────────────────────────────────┘
│ MCP · A2A · ACP · HTTP
│ payment proof
│ Agent-Authorization
┌───────────────────────────────────────▼──────────────────────────────────────┐
│ Agent Commerce Gateway (yours) │
│ │
│ protocol adapters → ExecutionPipeline │
│ │ │
│ ┌────────────────────┬──────────────┴──┬────────────────┐ │
│ ▼ ▼ ▼ ▼ │
│ AuthorizationProvider PaymentProvider BackendExecutor ReceiptStore │
│ (ap2) (x402, mpp) (bounded HTTP) (SQLite) │
└───────────────────────────────────────┬──────────────────────────────────────┘
│
┌──────────▼─────────┐
│ Your backend API │
└────────────────────┘
Every protocol adapter uses one execution pipeline, so payment enforcement does not depend on the adapter. See Architecture.
Configure a resource
resources:
market_report:
name: Premium Market Report
backend:
type: http
method: GET
url: ${MERCHANT_API_BASE_URL}/api/report
timeoutMs: 10000
pricing:
type: fixed
amount: "0.01"
currency: USDC
expose: [http, mcp]
payments: [x402]
The gateway calls your existing backend, which needs no SDK or changes.
npm run agent-commerce -- init # generate a config interactively
npm run agent-commerce -- validate # fails loudly, exits non-zero
Already have an OpenAPI description? Generate the resources from it (experimental):
agent-commerce import openapi ./openapi.yaml
It writes a reviewable resources: fragment with path, query and JSON body
bindings and supported schemas. The source document never decides price or
exposure: pricing appears only with --free and expose only with
--expose, and credentials are never imported. See
OpenAPI import for the supported subset.
Protocol support
| Protocol | Status | Pinned revision |
|---|---|---|
| HTTP | Supported | native routes |
| MCP | Supported | 2026-07-28 and 2025-11-25, @modelcontextprotocol/server@2.3.0 |
| x402 | Supported | x402 v2 (@x402/core, @x402/evm, @x402/svm, @x402/near), scheme exact, EVM and Solana; NEAR partial |
| MPP | Experimental | mppx@0.13.1, charge + evm + EIP-3009, USDC on Base Sepolia and Base |
| A2A | Experimental | A2A v1.0.1, binding JSONRPC, method SendMessage, x402 extension |
| ACP | Experimental | ACP 2026-04-17, REST checkout + discovery |
| AP2 | Experimental | AP2 v0.2.0, Direct Checkout Mandate verification |
| ERC-8409 | Experimental | draft SergeevDmitry/ERCs@0ee33230: signed quotes on x402 EVM offers |
| UCP | Planned | no implementation |
AP2 is an authorization method, not a transport. It verifies a mandate before a paid resource settles; the gateway does not issue the mandate or its Checkout Receipt.
The experimental entries above ship off by default with documented subsets:
A2A, ACP,
AP2, MPP and
ERC-8409 quotes. Supported
MCP and x402 are also scoped.
ACP exposes only its five checkout operations; merchant payment_data is never
converted into a gateway payment proof.
GET /.well-known/agent-commerce reports supportedSpec, capabilities and
unsupported for registered adapters and providers. agent-commerce doctor
prints the unsupported lists for A2A, ACP and AP2 when they are enabled.
Versioning
Minor releases may break experimental configurations, integrations or types:
MPP, A2A, ACP, AP2, ERC-8409 quotes, agent-commerce import openapi,
PaymentProvider.reconcile, the ReceiptStore methods and types added in
2.6.0, the idempotency resource settings with the store and provider methods
added in 2.7.0, and the ./testing entry.
The journal's Transaction type, ReceiptStore journal methods and
/api/transactions response shape are supported. The transactionId fields,
receiptCommitted and flags accepted by reconcile and resolve are also
supported.
Experimental runtime adapters and providers identify themselves with
status: experimental in GET /.well-known/agent-commerce. Pin experimental
integrations to an exact version and test upgrades before deployment.
Since 2.1, some supported x402 behavior and types
differ from 2.0; see Upgrading from 2.0 to 2.x.
Payment model
- Non-custodial. Buyer payments go directly to the merchant address in
payTo(recipientfor MPP). The gateway does not require buyer or merchant private keys. Local facilitator mode uses a separate key to pay gas. Experimental ERC-8409 quotes use a dedicated, unfunded signing key. - Fail closed. Missing, malformed, expired, replayed, wrong-amount, wrong-recipient, wrong-network and wrong-asset payments are rejected.
- Replay-safe twice over. The EIP-3009 or Permit2 nonce stops a double
spend on-chain; the gateway additionally reserves a
replayKeyderived from the authorization before it settles anything. - Retry after a lost response (experimental, disabled by default). With
idempotency.redelivery, buyers can retrieve a stored response without another charge using the same wallet, input and x402payment-identifieridor MPPIdempotency-Key. Matching exact resends skip verification until retention ends, an exception to fail closed. Responses are stored unencrypted in the receipt database.idempotency.merchantKeysends a derivedIdempotency-Keyto the merchant for paid operations. See Retry after a lost response. - On-chain settlement in CI. The end-to-end test asserts the buyer's balance falls and the merchant's rises by exactly the price, with a real transaction hash in the receipt. A log line saying "payment successful" would not count.
- One selected rail per resource. A resource can name several rails, such as
payments: [x402, mpp]; the first enabled one serves it, and nothing selects or falls back between rails per request. - Authorization is separate from payment. A resource can also require an AP2 mandate, verified before settlement and spendable exactly once. It never moves money and never unlocks a resource on its own - the payment still has to be real.
Public networks
Public networks use the same execution pipeline with a remote facilitator.
Choose the network in config; Base mainnet also requires allowMainnet: true.
It has actually settled
These transactions each moved 0.01 USDC from a buyer to a merchant through a remote facilitator:
| Rail | Network | Method | Transaction |
|---|---|---|---|
| x402 | Base Sepolia | EIP-3009 | 0xea41b234c4… |
| x402 | Base Sepolia | Permit2 | 0xa66f56cf39… |
| x402 | Base | EIP-3009 | 0x57ec81c2a3… |
| x402 | Base | Permit2 | 0x79d3da1145… |
| x402 | Solana Devnet | SPL transfer | 5289xbeyHJ… |
| x402 | Solana | SPL transfer | E7gAuG8pMA… |
| x402 | NEAR Testnet | NEP-141 transfer | 5tdtGLv4TB… |
| x402 | NEAR | NEP-141 transfer | 7hU3imHepV… |
| MPP | Base Sepolia | EIP-3009 | 0x5d9fcd111b… |
| MPP | Base | EIP-3009 | 0xf4255bd7ec… |
Facilitators broadcast these payments and paid their ETH, SOL or NEAR fees. Permit2 requires an initial, gas-paying token approval from the buyer. The NEAR runs used the suite's SDK facilitator and relayer; support remains partial pending hosted-facilitator settlement. Buyer and merchant keys stayed outside the gateway. The smoke suites verified settlement on-chain.
Reproduce with npm run test:testnet, which spends test USDC, or
npm run test:mainnet, which spends real USDC. Both run only by hand; see
Base mainnet and smoke tests. The
suites also cover NEAR testnet and mainnet (settled) and Arbitrum Sepolia (not
yet run).
The facilitator model
A facilitator verifies the buyer's authorization and broadcasts the transfer. It is the only component that needs gas.
facilitator.mode |
Who signs | Where it is allowed |
|---|---|---|
local |
this process, with a key you configure | a local Anvil chain; refused on mainnet, fails readiness elsewhere |
remote |
an HTTP facilitator you point at | anywhere |
With remote, the gateway holds no facilitator signing key. The buyer signs an
EIP-3009 authorization without ETH, and the facilitator pays gas. Because the
authorization fixes the recipient, amount and chain, the facilitator cannot
redirect the transfer. It can see authorizations and refuse service.
Remote facilitator auth supports none, bearer and cdp. CDP auth signs a
fresh JWT per request with the CDP API key. No auth mode needs an extra peer
package. Config load rejects other types.
Base Sepolia
payments:
x402:
enabled: true
network: eip155:84532
rpcUrl: https://base-sepolia-rpc.publicnode.com # health checks only
asset: "0x036CbD53842c5426634e7929541eC2318f3dCF7e" # Circle USDC
assetName: USDC
assetVersion: "2"
assetDecimals: 6
payTo: ${MERCHANT_WALLET}
maxTimeoutSeconds: 300
facilitator:
mode: remote
url: https://x402.org/facilitator
auth: { type: none }
Full config in examples/base-sepolia/. Test USDC
from faucet.circle.com; the buyer needs no ETH.
npm run test:testnet drives the whole flow and reads the result back off the
chain.
The local dev chain is eip155:31337, Anvil's default chain id, so the network
says whether a deployment is local, testnet or mainnet. doctor, health()
and /.well-known report it.
Agent Commerce Gateway MCP Fragen
Wie füge ich Agent Commerce Gateway MCP zu Claude Code hinzu?
Führe claude mcp add agent-commerce -- npx -y @devlab.group/agent-commerce in deinem Terminal aus. Füge --scope user hinzu, damit der Server in jedem Projekt verfügbar ist.
Wie füge ich Agent Commerce Gateway MCP zu Cursor hinzu?
Öffne ~/.cursor/mcp.json und füge den Server unter mcpServers hinzu, mit der Cursor-Config aus dem Installationsabschnitt oben.
Wie mache ich meine bestehende HTTP API für AI Agents verfügbar?
Du lässt das Gateway in deiner eigenen Infrastruktur laufen und beschreibst jeden Endpoint in YAML oder importierst eine OpenAPI-Spezifikation, um Entwürfe für die Resource-Definitionen zu generieren. Die Resources können dann über HTTP, MCP oder A2A bereitgestellt werden.
Speichert das Gateway meine Private Keys oder das Geld der Käufer?
Nein. Es ist non-custodial: Zahlungen gehen direkt an das Wallet des Merchants, und das Gateway benötigt weder den Private Key des Käufers noch den des Merchants.
Welche Payment-Protokolle unterstützt das Agent Commerce Gateway?
Bezahlte Resources nutzen x402 oder das experimentelle MPP und können zusätzlich ein AP2-Mandat erfordern, das beweist, dass der Käufer den Kauf genehmigt hat.
Welche Node-Version benötigt das Gateway?
Es wird Node >= 22 benötigt. Das Package enthält ein agent-commerce CLI mit Commands wie init, validate, start, doctor und import openapi.
Kann ich es ohne echtes Geld oder API Keys ausprobieren?
Ja. Der Quickstart nutzt Docker Compose, um einen temporären lokalen Stack mit Anvil, MockUSDC, einer Demo-Merchant-API, dem Gateway und einem Dashboard hochzufahren, und führt dann eine Demo aus, in der ein Agent etwas kauft.