There's been a lot of noise this month about AI agents being able to spend money. Stripe and Tempo shipped the Machine Payments Protocol in March, Cloudflare announced funded wallets with spending caps in August, and the x402 protocol quietly crossed nine figures of settled transactions. The pitch is simple: an agent hits an API, gets an HTTP 402 with a price, pays a few cents, and gets its answer. No signup form, no API key, no human in the loop.

That raised the question I couldn't find a good answer to: what can an agent actually buy today?

So I catalogued it. The Agent Directory is a curated index of live services that accept agent payments right now, over x402 or MPP, with real per-call prices. At launch it lists 16 services, with calls starting at a tenth of a cent.

Organized by what agents are missing

Most API directories sort by industry. That framing tells you nothing about why an autonomous agent would spend money, so I organized this one around the capability gaps that make an agent reach for its wallet.

Things agents can't see. Agents don't run JavaScript and browse from a single origin. Rendered pages, extraction, and search are the first things worth paying for.

Things agents can't know. Fresh, licensed, or niche data that isn't in any training set, priced per lookup instead of per subscription: live market data, government registries, change intelligence.

Things agents need to prove. This is the category I find most interesting. An agent's claim is worth more with a neutral receipt attached, so services selling verification per call (package security verdicts, registry lookups, calculations that cite their primary source) are quietly building the trust layer of the agent economy.

And two categories are deliberately listed as vacant: nobody yet sells agents waiting (durable timers, scheduled callbacks) or a way to be reached (an inbox for a confirmation code). An honest map shows the unclaimed territory. If you're building in this space, the empty rows are the opportunity.

The twist: this directory checks its own listings

Dead listings are how directories rot. Every "awesome list" starts useful and decays into a graveyard of parked domains, because nobody goes back to check.

I run an uptime monitoring company, so this directory checks. Every listing's discovery endpoint (usually its /.well-known/x402 payment manifest) is probed continuously by a Sentinel monitor, and the verification column on the page is live. When a service dies, the page says so before you waste a call on it, and chronic corpses get delisted.

Curation works the same way. While researching candidates, one "credible" service turned out to be a parked domain and another didn't exist at all. Nothing gets listed here without its endpoint answering first, which is why the directory is 16 services and not 60.

If you run an agent-payable API

If agents can discover and pay for your API today, it belongs here, and listing is free. Send the discovery URL through the contact form; if it verifies, it gets listed and monitored.

Not sure whether your API is agent-ready? The free AI Readiness Check now detects x402 payment manifests along with the rest of the agent-web signals, and hands you a starter manifest if you don't have one.

Where this goes

The interesting part isn't the list, it's the data underneath it. Those monitors are quietly building a record nobody else has: how reliable agent-payable services actually are over weeks and months, how fast the ecosystem grows, and how often services disappear. Give it a quarter and there will be numbers worth publishing.

For now: browse the directory, spend a tenth of a cent on something, and if you're building for agents, claim a vacant row.