This week a startup called Vaaya launched with a one-line pitch: "Turn your Github profile into Credit Card for your agents." Coinbase CEO Brian Armstrong amplified the launch on X. It matters less for the announcement itself than for the primitive it introduces — the first buy-now-pay-later for machine-to-machine payments.
Let me be precise about what Vaaya actually is, because the details are the story. It is a pay-per-call MCP server (npx @vaaya/mcp install) that fronts 1,400+ tools — Exa, Firecrawl, Fal, Apollo.io, Browserbase, Modal — with no API keys. Payment is the authentication. Billing rides x402 USDC on Base. Every call is priced before it runs, carries a max_cost_cents hard ceiling, and a failed call costs zero.
So far that is MCP aggregation, which other players already do. The genuinely new part is the credit line:
Managed wallet + credit line. Vaaya holds the wallet and issues a small line underwritten by your GitHub profile — "tens of dollars, not thousands."
Card-gated. Nothing from the line is spendable until a card is on file, and that card settles whatever gets drawn.
Repay history is the real signal. GitHub stars and account age carry the least weight because they're cheapest to fake. The maker's words: spend-and-repay history becomes "effectively a FICO equivalent for a developer profile."
By now x402 is a commodity — Coinbase, Circle, Cloudflare, and AWS all reference it. The friction that actually stops an agent from paying is upstream of the rails: how do you get spendable value into an agent's wallet without a human doing KYC, top-up, and key management? That is the step every early M2M project stumbles on.
Two answers now compete for the same scarce thing — that first funded wallet:
| minia2a (trial-first) | Vaaya (credit-first) | |
|---|---|---|
| Primitive | Grant — 15 free trial calls, 500 credits on registration | Credit — GitHub-underwritten line, card-gated |
| Wallet | Self-custody — agent brings its own key | Managed — Vaaya holds it |
| Repayment | None | Card settles what gets drawn |
| Trust signal | Verify-first + reliability scoring + honest volume | Spend-and-repay history ("agentic FICO") |
| Surface | Open marketplace (1,600+ services) | One curated MCP catalog |
These are not mirror images — they solve opposite halves of the funnel. Grant gets an agent to its first call with zero friction but caps spend at the free tier. Credit unlocks real paid spend but puts a human card back in the loop. Grant is the onboarding answer; credit is the activation answer.
The moat in Vaaya's model is not the tool catalog and not the rails. It's the claim that an agent's spend-and-repay history becomes a portable credit score for a developer. If that catches on, it is the kind of trust layer a self-custody marketplace cannot copy without becoming a custodian — a positioning trade-off, not an accident.
Two honest caveats before anyone over-rotates. First, the community challenge on launch day was the obvious one: rented or aged GitHub accounts can fake underwriting. Vaaya's answer — card-gating plus deliberately small limits — is sound but unproven at scale. Second, the whole category's real settled volume is still tiny; the scarce signal in M2M payments today is honest volume, not headline counts.
The takeaway is not that one model wins. It's that the agent payment stack just got a new layer — credit — sitting on top of the protocol and settlement layers that already commoditized. Whoever owns the trust primitive for that layer owns the next moat.