Agentic commerce · open standards

Add commerce to any agent.
Over the standards you'd pick anyway.

Paygent isn't a proprietary API — it's an agent that speaks the Universal Commerce Protocol: a discovery document, an agent card, and an MCP checkout endpoint. If your runtime can call a tool, it can sell, watch prices, optimize cards, and complete checkout.

All money and points math is deterministic and server-side — never delegated to a model.

embed_commerce.py
# 1 · Discover the agent over standard surfaces
card = fetch("https://shopchat.agent/.well-known/agent_card.json")
ucp  = fetch("https://shopchat.agent/.well-known/ucp")

# 2 · Open a session bound to YOUR wallet provider
session = paygent.session(
    user="usr_8f3a",
    wallet_provider="your-bank-or-program",   # pluggable
)

# 3 · Arm an autonomous purchase, bounded by a mandate
session.watch(
    product="wh-1000xm6",
    target="-15%",
    mandate={ "max_amount": 400.00,
              "scope": "this_cart",
              "expires": "7d" },
)

# → price hits · best card computed · points applied
# → checkout completes with a verifiable authorization
UCP · /.well-known/ucp Agent card · /.well-known/agent_card.json MCP checkout · /ucp/mcp UCP Loyalty Extension · end-to-end
Standard surfaces

Integrate the way the ecosystem already works.

No proprietary SDK required to start — the same surfaces the UCP ecosystem uses for discovery and checkout are how you talk to Paygent.

🪪

Agent card

A published, verifiable identity for the commerce agent — what it can do, how to reach it, and the key material to authenticate it. Your host runtime discovers capability the standard way.

🧭

UCP discovery

The /.well-known/ucp document declares checkout schemas, payment handlers, and the loyalty extension — machine-readable, versioned, and standard across the ecosystem.

🔧

MCP checkout tools

Create, update, complete, and cancel checkouts as MCP tool calls — idempotency keys preserved end-to-end, loyalty block on every response.

Why build on Paygent

The agent parts are easy. The money parts aren't.

Anyone can wire an LLM to a product feed. What you don't want to build twice:

⚖️

Deterministic money math

Totals, discounts, and point earning are computed by audited services and forced verbatim into agent output. The model decides what; it never invents how much.

🔐

Scoped spend mandates

Autonomous purchases carry a user-scoped, expiring, merchant-verifiable authorization — amount-bounded and cart-bounded. Cryptographic accountability, not a chat log.

🏦

Pluggable wallet & loyalty

Bind sessions to your wallet provider — cards, balances, points. The loyalty engine handles escrow, real-time redemption, and crash-safe capture behind one interface.

Autonomous flows, safely

Price watches complete checkouts while the user is away — idempotent, replay-protected, and bounded by the mandate they approved.

🧠

Model-agnostic runtime

Local or hosted models, swapped by configuration. The correctness lives in the services, so the agent works on a 8B model or a frontier one.

🧾

Loyalty block, spec-compliant

Every checkout response carries the UCP Loyalty Extension: earning forecast, per-rule breakdown, applied discounts — ready for your UI to render.

Integration paths

Three ways in, one engine underneath.

1

Call it as a tool

Your existing assistant delegates commerce intents to Paygent over MCP / agent-card discovery. Lightest touch.

2

Embed the SDKs

Drop the chat surface into your iOS or web app with reference SDKs — sessions, checkout widgets, wallet sync.

3

White-label the stack

Run the full engine under your brand with your catalog and your wallet provider behind the interface.

4

Go autonomous

Whichever path: arm price watches with mandates and let checkouts complete on your users' terms.

Early access

Build the commerce capability
you don't want to maintain.

We're onboarding a small number of design partners — host runtimes, platforms, and wallet providers — ahead of general availability.

Get API access Ask an integration question