Cashfree Gives AI Agents Six Tools To Complete Checkout

Checkout just got a tool belt. Cashfree Payments announced on November 6, 2026, that its Hosted Checkout now supports WebMCP, giving browser-based AI agents a structured way to handle key payment steps. The capability is live on redirect checkout and requires no merchant code changes.

Cashfree says its Checkout for Agents tools can cut the tokens needed to complete a checkout by more than 40%. The company also describes itself as the first payments company in India to bring WebMCP to checkout — a claim that arrives as retailers race to make AI shopping less conversational and more transactional.

Cashfree identified payment as a recurring obstacle in conversations with more than 50 founders building AI agents. On a conventional checkout page, an agent must interpret buttons and fields, simulate clicks, respond to layout changes, and cope with pop-ups; every extra step consumes tokens and creates another chance to stall or enter incorrect information.

Six tools replace guesswork with structured actions

WebMCP gives agents named tools that return structured results instead of forcing them to interpret the page visually. Cashfree’s checkout publishes six tools: get_checkout_summary, list_payment_methods, open_payment_method, back_to_payment_methods, show_upi_qr, and get_payment_status.

The order-summary tool returns the merchant name, currency, order amount, payable amount after offers, applied and available offers, and a masked phone number. The status tool reports not_started, pending, failed, or success — but only for the payment started on the current page.

Tool availability changes with the page and merchant settings, so agents must read the live tool list instead of assuming every option exists. The tools run inside the shopper’s own browser checkout session, without API keys or a separate server, and they cannot do anything the page itself cannot already do.

In the UPI flow, the agent reads the order, lists the payment methods enabled for it, and then requests a QR code. Cashfree marks show_upi_qr as consequential, allowing the browser or agent to ask for confirmation before running it; the shopper still scans the code with any UPI app and approves the payment with a UPI PIN.

The agent can then read the payment status. For other payment methods, it can list and open an option, but the shopper completes payment on the page.

Agents assist with payment but do not become payees

Human intervention remains necessary when checkout presents a convenience fee, a payment-route choice, autopay setup, or a decision about a pending previous payment. No tool independently moves money or accepts card numbers, CVV, one-time passwords, or payment tokens.

Cashfree also limits what the tools return: results omit email addresses and cart items, while merchant-supplied text such as offer descriptions is marked as untrusted data rather than instructions for the agent. Agent-assisted payments use the same checkout code, offers, fraud and risk checks, and error handling as shopper-initiated payments.

Merchants confirm completed payments through the same Get Order API or payment webhooks before fulfillment. In other words, the agent gets a narrower operating lane, not a magic wand. A sensible distinction, given how many systems still treat “automation” as permission to improvise.

Checkout for Agents is rolling out in phases, and browser support and available tools can change. That makes the live tool list part of the integration contract, rather than a decorative detail in documentation.

The timing reflects a wider push toward agentic commerce. In September, TRM Labs searched the x402 network for agent-initiated payments and found $52.7 million settled across 198.9 million transactions since May 2025. After excluding self-payments, bulk flows from one or two payers, and sellers with fewer than 10 buyers, $25.62 million remained.

TRM estimated that 0.6% to 7.5% of that remaining activity was plausibly initiated by an AI agent, depending on how strict the test was. A script and an agent leave the same record on-chain, so the real number could be higher — but it cannot be shown to be much higher. Most machine commerce still looks like ordinary automation wearing a more expensive label.

Retailers are preparing for more activity regardless. Amazon added more than 100 features to Alexa for Shopping, including shopping lists, gift registries, product subscriptions, and payment and billing questions; the tool is available through the Amazon Shopping app without Amazon devices or a Prime membership.

Rajiv Mehta said, “This is just the beginning,” adding, “We’re continuing to invent new ways to save customers time and make shopping even more personal.” Target added AI review insights and image-based product recommendations to its shopping app, while Walmart announced AI agents that can shop and make purchases for customers.

Consumer behavior is moving in the same direction. Omnisend found that 38 percent of U.S. shoppers made purchases directly through ChatGPT in 2025, while 63 percent used AI to research purchases, find products, or create gift lists. Adobe found that 77 percent of consumers said AI made them feel more confident about online purchases in a July survey of 5,000 people.

Adobe projects $275 billion in online spending from November to December, nearly $20 billion more than shoppers spent in 2025, alongside more than $1 trillion in U.S. holiday retail sales. Traffic from shoppers clicking AI-provided links is projected to rise 130 percent in November and December.

Secure Technology Alliance and Merchant Advisory Group are also working through the Agentic Trust and Commerce Forum to align payments and identity stakeholders around how agents initiate, authenticate, and complete transactions. Cashfree’s approach is narrower but more practical: give the browser defined actions, require confirmation where the action matters, and keep the shopper in control.

Stay connected via Google News

Follow us for the latest AI News