Skip to main content
An agent does the work. When that work reaches money, Payman executes it against a person’s own accounts: with their consent, under their limits, and with their approval on anything that moves money. They connect their Payman account once, on Payman’s own pages. It holds whichever accounts they use: a bank, a corporate card, a crypto account, a shopping wallet. Nobody else ever holds a credential, and nobody has to build an integration with any of them.

Two ways in

I am building an app

Your product, your customers, your UI. Install the SDK, hold one key, and call Payman for the money step.

I want to use it myself

Payman inside Claude, ChatGPT, Grok or Cursor, against your own accounts. Nothing to build.
They are the same money surface with two front doors. Same consent page, same approvals, same accounts.

Building an app

Payman Connect is the execution layer, not the brain. Your agent decides that an invoice should be paid or a balance checked, calls Payman for that one step, and reads the answer back in the same turn. You hold a pointer to your customer’s consent, never a credential. The SDK carries the rest: one secret, PAYMAN_APP_KEY, and one call per action, with the consent handshake, the encrypted connection store, the approval flow and the replay all shipping with it.

Quickstart

A working integration in one page.

Add it as a tool

Your agent already runs a model. This is the one tool it needs.

How it works

The loop end to end.

The Wallet

One customer, several accounts, and who picks between them.

Using it yourself

No app, no key, no code. Give your AI assistant one address and it connects to your own Payman account: a URL for hosted chats like Claude and ChatGPT, a one-line command for Claude Desktop, Claude Code or Cursor. You sign in on Payman’s own pages, your bank credential goes on Payman’s own pages, and anything that moves money still stops for your approval there.

MCP server

Both ways in, start to finish.