Project showcase — Grantex

I built the payment layer for AI agents.

Grantex is agentic payment infrastructure. An AI agent calls one API, receives a single-use virtual card with network-enforced spending controls, completes a purchase, and the card ceases to exist. No checkout. No human in the loop.

< 300ms

Card creation to credential

Zero

Human touchpoints per transaction

MCP

Any agent framework, one line of config


grantex.ai / agent

Reorder toner cartridges — stock below 10 units
--:-- Waiting for task...
Transaction receipt Approved
MerchantStaples Business
Amount$47.80
CategoryOffice supplies
Card4000 •••• •••• 0021 (expired)
Limit enforced$100 max per authorization
Card statusDestroyed after use

01

Agent receives task

02

Grantex API called

03

Virtual card issued

04

Purchase executed

05

Card destroyed

06

Receipt logged


01 — Policy

Rules baked into the credential

Spending controls are encoded at the payment network level. Merchant category locks, amount caps, and single-use expiry are enforced by the network itself. Not software instructions an agent can reason around.

02 — Credential

Disposable by design

Each transaction generates a fresh virtual card. Created for one purpose, used once, and then ceases to exist. No standing credential that can be extracted, reused, or compromised after the fact.

03 — Protocol

Native to any agent

An MCP server means any agent framework can call Grantex with one line of configuration. Claude, LangChain, AutoGPT. No SDK, no custom integration. Payment becomes a native agent capability.


"We had to build this from scratch because it does not exist natively in most issuer products today. The agent cannot work with a credential it cannot interrogate. A static card number with no machine-readable parameters is invisible to an agent making a routing decision."

From “I built agentic payments infrastructure. Here is what nobody is talking about.” — Sean McAuliffe


A working prototype, built to think with.

Grantex is not a company. It's a credibility build — vision to running prototype in a handful of weekends, using Claude Code, Lovable, Supabase, and GitHub. I built it to pressure-test a thesis I'd been writing about: AI agents need payment credentials they can interrogate, not static card numbers. The essay that came out of the build is below.

If you've got a fintech idea you need rendered as a working prototype — fast, by someone who knows the domain — that's what I do. The Grantex build is what it looks like.