Your AI agent needs to pay for things. API calls, data feeds, inference — the agent economy runs on microtransactions. But agents can't open bank accounts, can't pass KYC, and can't type credit card numbers. They need agent-native wallets.
In 2026, five distinct approaches have emerged. Here's how they compare — who they're for, what they cost, and which one fits your stack.
| Solution | Custody | Chain | Min Payment | Dev Experience | Status |
|---|---|---|---|---|---|
| Circle Agent Stack | Custodial (policy-guarded) | 34+ EVM + Solana | $0.000001 | SDK + CLI | GA (May 2026) |
| Coinbase AgentCore | Custodial (AWS-managed) | Base | ~$0.01 | AWS Bedrock SDK | GA (2026) |
| Franklin (BlockRun) | Non-custodial | Base | $0.01 | TypeScript SDK | Open source |
| agenticpay | Non-custodial | Solana | ~$0.001 | MCP server | Open source |
| minia2a Marketplace | Platform credits → USDC | Base | $0.01 | Zero-setup REST | Live (207 services) |
Circle launched Agent Stack in May 2026 as the first full-stack financial infrastructure where agents are the customers. Four products:
Teams building agent fleets that need programmatic custody with compliance guardrails. If you're deploying dozens of agents with spending limits, approval chains, and audit trails — this is your tool.
It's an enterprise product. Circle's ecosystem is curated and gated. You don't just deploy an agent — you integrate with Circle's full stack. Great for funded startups; heavy for indie builders.
// Circle Agent Stack — programmatic wallet with policy
const wallet = await circle.agentWallet.create({
agent: "research-bot-7",
policy: { maxPerTx: "5.00", dailyCap: "50.00", allowedServices: ["*"] }
});
await wallet.pay("0x...", "0.50"); // $0.50 USDC, gas-freeCoinbase partnered with AWS to bake x402 payments directly into Amazon Bedrock's AgentCore Payments. Agents discover services, call them, and pay in USDC — all within the AWS ecosystem. Settlement in ~200ms at sub-cent cost.
Backed by the x402 protocol (HTTP 402 Payment Required), with support from Google and Visa. 165 million transactions, 69,000+ agents, ~$50 million cumulative settlement.
Teams already on AWS Bedrock building agent workflows. If your agent lives in the Amazon ecosystem, this is the path of least resistance.
Vendor lock-in. Your agent's payment capability is tied to AWS. If you ever want to run outside Bedrock, you're rebuilding the payment layer. Also: AWS pricing complexity on top of payment fees.
Franklin by BlockRun is an open-source TypeScript agent that autonomously spends USDC across 55+ API and model providers. It introduced YOPO (You Only Pay Outcome): provider cost + 5%, settled per action.
No subscriptions. No API keys. Wallet address is identity. Apache 2.0 license.
// Franklin — wallet address IS the API key
import { Franklin } from '@blockrun/franklin';
const agent = new Franklin({ wallet: process.env.PRIVATE_KEY });
const result = await agent.call('web-scrape', { url: 'https://example.com' });
// Pays ~$0.01 USDC automatically. No key exchange. No signup.Developers who want full control. Open-source, self-hosted, no custody. If you're comfortable managing private keys and want zero platform dependency — Franklin is the answer.
You manage the keys. Lose the key, lose the funds. Also: 55 providers is a curated set — you can't just call any API on the internet.
agenticpay brings payments into the MCP (Model Context Protocol) layer. Any tool declares a price; any agent pays it in USDC. Built on Solana with ~1.5s settlement.
The key insight: payments should be a tool primitive, not a separate integration. Define price in tool metadata, handle settlement in the protocol layer.
// agenticpay — price as tool metadata
// price declared in MCP tool definition
{ "name": "deep-research", "price": "0.05", "currency": "USDC" }
// Agent calls tool → MCP server handles x402 flow → agent wallet paysMCP ecosystem developers. If your agents already use MCP servers, agenticpay adds payments without changing the tool-calling pattern.
Early stage. Solana-only. MCP-only — agents outside that ecosystem need a different solution.
minia2a takes a different approach: instead of giving your agent a wallet, it creates a market where agents discover and pay for services without setup.
No API keys. No registration. No KYC. Services list themselves. Agents get free trial credits (currently 500 per new agent). When credits run out, agents pay in USDC via x402 on Base.
Key numbers (live, August 2026):
# minia2a — zero setup, agent pays per call
curl -X POST https://minia2a.uk/api/x402/web-scrape \
-H "Content-Type: application/json" \
-H "x402-payment: ..." \
-d '{"url":"https://example.com"}'
# No API key. No signup. Just pay and get data.Agent developers who want to use paid services without managing wallets themselves. Also: API providers who want to monetize without building auth, billing, and rate-limiting from scratch.
Young marketplace. 207 services but many are utility endpoints — not every category is deep. 35 wallet users means the paying market is small (though growing). You're betting on network effects.
The honest answer: it depends on what you're building.
The wallet problem isn't solved yet — not really. Here's what the numbers tell us:
The bottleneck isn't the technology of wallets. It's three things:
The wallet infrastructure exists. The missing piece is agent-native onboarding — where an agent can create a wallet, fund it, discover services, and pay for them without a human in the loop. That's the real prize.