> ## Documentation Index
> Fetch the complete documentation index at: https://developers.t2000.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Mail a physical card

> Agent writes the message, generates the front, and ships a real postcard via USPS. ~$2.08 end to end.

The most interesting demo of agentic payments isn't another LLM chain — it's letting an agent reach into the physical world. This recipe writes a warm note with Claude, renders the card front with fal.ai, and prints + mails it via [Lob](https://lob.com) at USPS rates.

You pay \~\$2.08 in USDC. A physical postcard shows up in someone's mailbox 3–5 days later.

<Note>
  The card front is the only part a capable client could draw itself; the **un-fakeable** part is the last step — a sandbox can't put a stamped postcard in a real mailbox. That's the durable value of the gateway: a real-world action settled in USDC.
</Note>

***

## The prompt

```
Use t2 services. Write a warm note for {recipient}, render it as a card front via fal.ai, and mail it to {address} via Lob.
```

***

## What runs

1. `POST /anthropic/v1/messages` — Claude writes the note text (\~\$0.02)
2. `POST /fal/fal-ai/flux/dev` — Flux renders the card front (\~\$0.06)
3. `POST /lob/v1/postcards` — Lob prints + mails the physical postcard (\~\$2.00)

***

## Run it

### SDK

```typescript theme={null}
import { T2000 } from '@t2000/sdk';

const agent = await T2000.create();

const occasion = 'birthday';

// Lob needs a STRUCTURED address (line1 / city / state / zip / country),
// not one crammed string — a single `address_line1` fails validation.
const recipient = {
  name: 'Mum',
  address_line1: '210 King St',
  address_city: 'San Francisco',
  address_state: 'CA',
  address_zip: '94107',
  address_country: 'US',
};

const note = await agent.pay({
  url: 'https://mpp.t2000.ai/anthropic/v1/messages',
  method: 'POST',
  headers: { 'anthropic-version': '2023-06-01' },
  body: JSON.stringify({
    model: 'claude-sonnet-4-5',
    max_tokens: 200,
    messages: [{
      role: 'user',
      content: `Write a warm ${occasion} note from me to ${recipient.name}. 3-4 sentences. Personal, not gushing.`,
    }],
  }),
});

const cardFront = await agent.pay({
  url: 'https://mpp.t2000.ai/fal/fal-ai/flux/dev',
  method: 'POST',
  body: JSON.stringify({
    prompt: `Painterly ${occasion} card front. Warm palette, illustrative, no text.`,
    image_size: 'landscape_16_9',
  }),
});

const noteText = (note.body as { content: { text: string }[] }).content[0].text;
const imageUrl = (cardFront.body as { images: { url: string }[] }).images[0].url;

const postcard = await agent.pay({
  url: 'https://mpp.t2000.ai/lob/v1/postcards',
  method: 'POST',
  body: JSON.stringify({
    description: `${occasion} card to ${recipient.name}`,
    to: recipient,
    from: {
      name: 'Phillip',
      address_line1: '500 Market St',
      address_city: 'San Francisco',
      address_state: 'CA',
      address_zip: '94105',
      address_country: 'US',
    },
    front: imageUrl,
    back: noteText,
    size: '4x6',
  }),
});
```

### Claude Desktop (MCP)

The chain above runs end-to-end in Claude Desktop if you paste the natural-language prompt — the MCP wallet handles the three 402s in sequence and Claude orchestrates the data flow.

***

## Expected output

```
3 calls · ~$2.08 · 0 taps · queued for USPS delivery in 3–5 days
```

Lob returns a tracking ID + expected delivery date.

***

## Extend it

* Use **Lob `/v1/letters`** (\~\$3.00) for multi-page letters instead of postcards
* Verify the destination first with **Lob `/v1/verify`** (\~\$0.02) to catch typos before paying \$2.00
* Swap Flux Dev for **Recraft 20B** (`/fal/fal-ai/recraft-20b`) for vector-style illustrations
* Generate batches: 10 unique cards to 10 friends, \~\$20.80 total, one tap-free pass