> ## Documentation Index > Fetch the complete documentation index at: https://docs.prava.space/llms.txt > Use this file to discover all available pages before exploring further. # Use Cases > What people build with Prava, shown as concrete flows. Prava does one thing: it lets an AI agent pay with a real card, without ever holding that card. Here is what that looks like in practice. ## An agent buys something for you You tell your agent "order me a phone case under \$20". The agent, connected via [MCP](/mcp/overview) (Model Context Protocol) or the [CLI](/prava-pay/overview): 1. Searches products across merchants and picks a case: \$18.50. 2. Locks the live price with a quote. 3. Sends you a payment link. You open it and approve with Face ID. 4. The agent completes the checkout. You get an order id; the agent never saw your card. If the total had changed mid-checkout, the charge would not have gone through. The credential is locked to that merchant and that amount. **Start here:** [Connect via MCP](/mcp/connect) or [Prava Pay quickstart](/prava-pay/quickstart). ## An agent pays a bill you already know No product discovery needed. You say "pay my $54.20 internet bill at Comcast". The agent creates a payment session for exactly $54.20, you approve with a passkey, and it pays with the one-time credentials. Wrong-amount or wrong-merchant use is declined. **Start here:** [Payment sessions (CLI)](/prava-pay/sessions) or [create\_payment\_session (MCP)](/mcp/tools#create-payment-session--checkoutrun). ## Your AI app takes payments from its users You're building a shopping assistant, a travel planner, or a booking bot. Your users need to pay *inside your product*. You integrate the [SDK + API](/sdk/overview): your backend creates a session, your frontend shows Prava's secure card form, and your app receives one-time credentials to complete the purchase. Your servers never touch a card number, which keeps you out of PCI scope (the security compliance that applies once you handle card data). Example: a stylist app's user picks a $79 jacket. The app creates a session for $79, the user enters their card and approves in the embedded form, the app checks out at the boutique with the one-time token, then reports the result. **Start here:** [Add Payments to Your AI App](/guides/add-payments-to-your-ai-app). ## Human-in-the-loop, by design Every flow above has the same shape: **the agent proposes, the human approves, Prava executes.** The approval is a passkey (Touch ID / Face ID) on the user's own device. An agent cannot skip it, and an approval for one purchase authorizes nothing else. ## Recurring payments (planned) A mandate is the user's recorded permission to pay. Today every mandate is one-time: one approval, one purchase. Recurring mandates are planned. Once shipped, a user will approve a subscription once (say, \$12/month at one merchant), and the agent can invoke it monthly within those limits, without a fresh passkey each time. Until then, each payment needs its own approval. ## Where to next Agent, embedded, or hosted. Decided in 10 seconds. The lifecycle behind every flow above. For merchants: what works with zero changes.