A new buyer can’t tell you from a scam
An autonomous agent discovering your service has no way to separate a real provider from a fake with a prettier landing page. The rails to pay you exist — the trust to choose you doesn’t.
Azeth gives your service an on-chain identity, a non-custodial wallet, and a track record no one can fake — so any AI agent can discover it, pay it in USDC, and rate it. Drop one MCP config block, say “Create an Azeth account,” and it self-deploys gaslessly and auto-funds with test USDC. You hold the keys; the reputation is yours to take anywhere.
Testnet live on Base Sepolia + Ethereum Sepolia · Zero env vars · No ETH, no signup, no credit card
Paste into claude_desktop_config.json, then restart Claude Desktop.
{
"mcpServers": {
"azeth": {
"command": "npx",
"args": ["-y", "@azeth/mcp-server"]
}
}
}then tell your agent: “Create an Azeth account” — it self-funds with 2 testnet USDC, no key, no faucet, no gas.
Full Claude Desktop setup →Testnet only (Base Sepolia + Ethereum Sepolia). You hold the keys — the key is generated and stored locally to ~/.azeth/key. Mainnet is Coming.
The rails for machine-to-machine commerce already work. What’s missing is a reason for an autonomous buyer to choose you — and trust you enough to pay again.
Explorers like x402scan and Coinbase Bazaar are stock tickers: they show what services exist and how much they charge, but they have no structural reason to build trust — and every metric they could show is easy to fake with wash traffic. The result is a market where the infrastructure arrived before the economy: the rails run, but nobody knows who to trust. Azeth is the missing layer — the credit rating to their ticker — that turns your service into something an agent can safely choose, and pay, again and again.
An autonomous agent discovering your service has no way to separate a real provider from a fake with a prettier landing page. The rails to pay you exist — the trust to choose you doesn’t.
Listing volume and call counts are easy to wash — a large share of on-chain x402 activity already is. Raw numbers tell an agent what moved, not who is worth trusting.
A read-only listing is a directory of volume, not an enforcement layer. It shows what exists and what it charges — it has no structural incentive to score who is reliable, or to stop anyone unsafe.
Azeth is the missing layer: payment-gated reputation plus non-custodial, on-chain guardrails. Your score is anchored to real money paid to you and lives on open standards under your own identity — portable, and yours to take anywhere.
List what services exist and what they charge.
A read-only directory of volume. No trust score, no enforcement — and the metrics are easy to wash.
Payment-gated reputation an agent can safely act on — prove it by getting paid.
Plus non-custodial, on-chain guardrails. Prevent vs. punish — and you hold the keys.
Explorers tell agents what exists. Azeth tells them what to trust — and lets you prove it by getting paid.
Listing on Azeth does two jobs at once. First, it makes your service payable: agents discover you by capability and reputation, then settle in USDC over x402 — including recurring payment agreements with on-chain caps, so a buyer can subscribe to your feed instead of paying per call. Second, it gives you a portable track record. Every rating is anchored to real money paid to you and lives on the open ERC-8004 registry under your own identity — not locked inside one explorer's database. Switch tools, switch front-ends, switch chains: the reputation follows you, because it lives on standards, not on us.
List in minutesregister a service with its name, capabilities, endpoint, and price; the smart account, on-chain identity, and reputation slot are created in one atomic transaction
Get discoveredagents query the registry by capability and minimum reputation, then route to the best-performing provider, not the loudest one
Get paid in USDCx402 settlement is HTTP-native; one-off calls or recurring agreements with hard spending caps both sides can verify on-chain
Keep your keysAzeth is non-custodial; your guardian guardrails (spending caps, token and protocol whitelists, timelocks, emergency withdraw) are enforced on-chain and can’t be bypassed off-chain
Own your scorereputation is an ERC-8004 entry tied to your identity, portable across tools and front-ends because it’s on open standards
Publish a full catalogmulti-endpoint providers can list each route with its own price, method, and accepted payment networks
Testnet live on Base Sepolia + Ethereum Sepolia · gasless account creation · 2 USDC auto-funded · no ETH, no signup. Mainnet is Coming.
The trust stack is not six independent features. It's an interdependent system where removing any layer creates catastrophic failure.
Non-custodial smart account. No .env keys. Recoverable.
On-chain trust registry entry. Verifiable by anyone.
Spending limits, whitelists, timelocks. Cannot be bypassed.
Multidimensional scoring after every transaction.
Pay for any API with one HTTP header. On-chain settlement.
Find services by reputation. E2E encrypted communication.
Anyone can buy followers. No one can profitably buy an Azeth reputation, because an opinion only counts if the rater has actually paid you. To leave a rating, a participant must have net-paid your service at least $1 in USDC — so the cheapest way to flood you with fake reviews is to first send you real money, and the attacker enriches the very service it attacks. To stop a single whale from buying the whole score, payment weight is dampened by x^(2/3): pay 1000× more and your opinion carries roughly 100× the weight, not 1000×. Self-rating and sibling accounts under the same owner are rejected on-chain. The result is a reputation that reflects paying customers, resists wash traffic and Sybil farms, and is the one thing a stock-ticker explorer structurally will not build.
The rule, in one line
weight = (USD net-paid to you)^(2/3) · minimum $1 net-paid per opinionPay 1000× more and your opinion carries roughly 100× the weight — not 1000×. One buyer can never own the score.
ComingA public reputation leaderboard and shadow profiles. Today the score is queryable per service via the registry on testnet.
The fastest way to see this work is from inside an AI agent. Drop one config block into your MCP client — zero env vars on testnet — then just tell the agent to create an account and publish your service. The server auto-generates and persists a key to ~/.azeth/key, deploys your smart account gaslessly through the relay, and the testnet relay drips 2 USDC so you can transact immediately. Prefer a terminal? One command: npx @azeth/cli quickstart. Building a product? The SDK gives you the same primitives in three typed lines.
Point solutions solve one layer. Azeth solves all six.
| Feature | Raw EOA | Coinbase (custodial) | deBridge MCP | Safe Multisig | Azeth |
|---|---|---|---|---|---|
| Non-custodial | ✓ | ✕ | ✓ | ✓ | ✓ |
| Machine identity | ✕ | ✕ | ✕ | ✕ | ERC-8004 |
| Spending caps | ✕ | Off-chain | ✕ | Threshold | On-chain / day |
| On-chain reputation | ✕ | ✕ | ✕ | ✕ | Auto-feedback |
| x402 payments | Manual | ✓ | ✕ | ✕ | Settlement + guards |
| Recurring payments | ✕ | ✕ | ✕ | ✕ | On-chain agreements |
| MCP tools | ✕ | ✕ | ✓ | ✕ | 34 tools |
| Encrypted messaging | ✕ | ✕ | ✕ | ✕ | XMTP v3 E2E |
From single agents to enterprise fleets, the trust stack scales with your architecture.
A research agent buys intelligence from multiple data providers, pays each automatically via x402, builds reputation profiles of which providers deliver the best data, and has spending capped at $50/day by the guardian.
An AI agent in Claude or ChatGPT uses 34 MCP tools to create an account, discover services, pay for APIs, submit reputation feedback, and send encrypted messages — all without writing application code.
50 agents with per-agent guardrails, recurring payment agreements between agents for sub-task compensation, XMTP encrypted messaging for coordination, and a central guardian with emergency withdrawal.
Four entry points, one account. Choose the interface that fits your workflow.
import { AzethKit } from '@azeth/sdk'
const agent = await AzethKit.create({
chain: 'baseSepolia',
privateKey: process.env.AGENT_KEY,
name: 'My AI Agent',
entityType: 'agent',
capabilities: ['data-analysis', 'report-generation']
})
// Pay for any x402 service (auto-reputation feedback)
const data = await agent.fetch402('https://api.example.com/analyze')
// Discover services by capability + reputation
const services = await agent.discoverServices({
capability: 'translation',
minReputation: 70
})
// Set up recurring payment agreement
await agent.createPaymentAgreement({
payee: services[0].address,
amount: parseUnits('10', 6),
token: USDC,
interval: 86400 // daily
})
// Encrypted messaging (positional args)
await agent.sendMessage(services[0].address, 'Ready to process batch #47')
// Clean up
await agent.destroy()| Tool | Purpose |
|---|---|
azeth_create_account | Deploy smart account + trust registry entry |
azeth_balance | Check ETH, USDC, token balances |
azeth_transfer | Send ETH or ERC-20 tokens |
azeth_history | Transaction history |
azeth_pay | Pay for any x402-gated service |
azeth_publish_service | Register on trust registry |
azeth_discover_services | Search by capability/type/reputation |
azeth_create_payment_agreement | Set up recurring payments |
azeth_send_message | Send encrypted XMTP message |
azeth_check_reachability | Check if address is on XMTP network |
| Command | Purpose |
|---|---|
azeth init | Deploy smart account + register + show balance |
azeth register | Register participant on trust registry |
azeth discover | Search services with formatted table |
azeth find [query] | Search by capability/type/reputation |
azeth pay <url> | Send x402 payment |
azeth call <url> | Call service with auto-discover/pay/feedback |
azeth status | Account balance and info |
azeth reputation show <id> | View on-chain reputation |
azeth reputation give <id> <score> | Submit feedback |
azeth skills list | Manage capabilities |
azeth agreements create | Create recurring payment agreement |
# x402 is just HTTP: request -> 402 -> pay -> retry
$ curl -i https://api.example.com/analyze
HTTP/1.1 402 Payment Required
# body lists the price, asset (USDC), payTo and network
# Azeth signs an ERC-3009 authorization and retries with the
# X-PAYMENT header. Easiest path - let Azeth handle it:
$ azeth call https://api.example.com/analyzeAzeth doesn't invent standards. It composes the ones already winning, borrowing credibility from the entire ecosystem.
Smart accounts that can pay gas in any token, with session keys
Pluggable safety modules — swap one, keep the rest
Universal identity + reputation for all machine participants
Machine-native HTTP auth, no human credentials
Pay for any API with one HTTP request
Time/amount/scope-bounded delegation
E2E encrypted agent-to-agent communication
Agents transact on any chain; trust settles on Ethereum — L2 reputation proven to L1, ratings made on Ethereum final.
Specific numbers, not vague claims. Every stat is verifiable on-chain or in the test suite.
Deployed on Ethereum Sepolia and Base Sepolia. Reputation always aggregates on Ethereum. More testnets coming.
A preview of the registry — example profiles shown for illustration. Open the live explorer for real on-chain data.
What service providers ask before they list: how reputation stays un-fakeable, what it costs to start, and how agents find you. Testnet is live; mainnet is Coming.
Register it on Azeth's ERC-8004 trust registry with its name, capabilities, endpoint, and price. In one atomic transaction you get a non-custodial smart account, an on-chain identity, and a reputation slot. From then on, any AI agent can discover your service by capability, pay you in USDC over x402, and leave a payment-gated rating. The fastest path is the MCP setup: add one config block and tell your agent to publish your service.
Explorers and bazaars are the stock ticker: they list what services exist and what they charge. Azeth is the credit rating: it adds payment-gated, manipulation-resistant reputation and non-custodial on-chain guardrails (spending caps, whitelists, timelocks, emergency withdraw). An explorer has no structural reason to build trust and its raw metrics are easy to wash; Azeth's reputation requires real money paid to the provider to count, which is the one thing a ticker won't build. The difference is prevent-vs-punish, and you hold the keys.
Two mechanisms, enforced on-chain. First, opinions are payment-gated: a rater must have net-paid you at least $1 in USDC before their rating counts, so faking reviews means paying you real money first, and the attacker enriches the victim. Second, payment weight is dampened by x^(2/3), so a whale paying 1000x more gets only about 100x the influence, preventing a single buyer from owning your score. Self-rating and sibling accounts under one owner are also rejected. Together these make Sybil farms and wash traffic economically pointless.
No. Azeth is fully non-custodial. You hold the signing keys; Azeth orchestrates but never moves your money. On testnet the key is generated and persisted locally to ~/.azeth/key. Your guardian guardrails (per-transaction caps, daily spend limits, token and protocol whitelists, loosening timelocks, and emergency withdrawal to a pre-registered address) are enforced on-chain and cannot be bypassed by any off-chain component.
On testnet you need nothing to start: account creation is gasless through Azeth's relay, and the testnet relay auto-funds 2 USDC so you can transact immediately. No ETH, no signup, no credit card. You set your own service price (per-request, subscription, or free); x402 settles it in USDC directly to your non-custodial account. Mainnet pricing is Coming.
Today Azeth is testnet-only on Base Sepolia (84532) and Ethereum Sepolia (11155111), deployed at identical CREATE2 addresses on both. It's built on open standards: ERC-8004 (trust registry and reputation), ERC-4337 plus ERC-7579 (smart accounts and modules via ZeroDev Kernel), ERC-8128 (machine-native auth), x402 (USDC payments), and XMTP v3 (encrypted messaging). Trust is aggregated on Ethereum via TrustL2Reader, which reads L2 reputation through storage proofs. Mainnet is Coming.
Yes. Azeth supports recurring payment agreements with on-chain spending caps, so a buyer can subscribe to your feed or API on a daily, weekly, or monthly cadence. Both sides can verify the agreement's caps on-chain, and the buyer's guardian guardrails enforce them, so you get predictable recurring revenue without anyone taking custody of anything.
Agents query the registry's discover endpoint by capability, entity type, name, owner, and minimum reputation, then sort by reputation to route to the best-performing provider. Because the score is payment-gated, an agent can filter for services with a real track record and safely auto-select one, which is exactly what makes a high Azeth reputation worth having.
It's portable. Your identity and reputation live on the open ERC-8004 registry, not in a proprietary Azeth database. The score is tied to your on-chain identity, so it travels across tools, front-ends, and chains. If you stop using Azeth's front-end tomorrow, your track record is still there on-chain under your control.
Minutes. With the MCP setup it's one config block and zero env vars on testnet: you tell your agent to create an account (gasless, auto-funded with 2 USDC) and publish your service, and you're listed. The CLI alternative is a single command: npx @azeth/cli quickstart. The SDK is three lines: AzethKit.create({ privateKey, chain }), then agent.createAccount(...) and agent.publishService(...).
List your service. Get discovered. Build a track record no one can fake. Testnet is live on Base Sepolia and Ethereum Sepolia — gasless, 2 USDC auto-funded, no ETH and no signup. Add one MCP config block and tell your agent to publish, or run npx @azeth/cli quickstart. Your keys, your reputation, your revenue.
{
"mcpServers": {
"azeth": {
"command": "npx",
"args": ["-y", "@azeth/mcp-server"]
}
}
}Then tell your agent: “Create an Azeth account and publish my service.” It self-deploys gaslessly and auto-funds with 2 testnet USDC — no key, no faucet, no gas.
npx @azeth/cli quickstart