Three ways an AI agent can buy physical goods in 2026

The short answer

Try the third way now, with no key

The plant request from the houseplants guide: pots made anywhere, sent in lots to a nursery in the US that pots and ships each customer's order. No key needed:

curl -sX POST https://api.etabli.io/v1/quotes/preview \  -H "Content-Type: application/json" \  -d '{"request":"Natural pastel self-watering pots in our own design, made cheaply in lots of up to 100 (overseas is fine) and sent to a nursery in the US. For each customer order, the nursery pots one of its houseplants in our pot and ships it to the customer within 2 business days. Up to $3 a pot and $14 an order, postage on top."}'

const response = await fetch("https://api.etabli.io/v1/quotes/preview", {  method: "POST",  headers: { "Content-Type": "application/json" },  body: JSON.stringify({    request: "Natural pastel self-watering pots in our own design, made cheaply in lots of up to 100 (overseas is fine) and sent to a nursery in the US. For each customer order, the nursery pots one of its houseplants in our pot and ships it to the customer within 2 business days. Up to $3 a pot and $14 an order, postage on top.",  }),});const plan = await response.json();for (const leg of plan.legs) {  console.log(leg.key, leg.title, `${leg.questions.length} open questions`);}

import requests plan = requests.post(    "https://api.etabli.io/v1/quotes/preview",    json={"request": "Natural pastel self-watering pots in our own design, made cheaply in lots of up to 100 (overseas is fine) and sent to a nursery in the US. For each customer order, the nursery pots one of its houseplants in our pot and ships it to the customer within 2 business days. Up to $3 a pot and $14 an order, postage on top."},    timeout=120,).json()for leg in plan["legs"]:    print(leg["key"], leg["title"], len(leg["questions"]), "open questions")

What the live API returned on 9 October 2026, in short (the houseplants guide shows all of it):

1. Buy from a catalog

The item exists and has a price; the API orders it. Print-on-demand services (Printful, Printify) put your design on their blank products, then print and ship each order. Gifting APIs (Goody) send a gift from their catalog, which the recipient accepts and addresses themselves. For things already sold in stores, agentic checkout protocols (OpenAI and Stripe's ACP, Google's UCP) let an agent buy from a merchant directly.

Use it when

• The catalog has the item.
• You want the price before you order.
• Orders come one at a time, as they happen.

Its limits

Only what's in the catalog. Your own product, your packaging, or a job that takes two workshops isn't there.

2. Source from a marketplace

A marketplace such as Alibaba.com lists factories and trading companies, many of them overseas. You search listings or post a request for quotation, message sellers, order samples, and agree minimums, payment and freight. It is built for people doing that work.

Use it when

• You need a standard product, in volume.
• A person will run the sourcing: compare quotes, check samples, arrange freight and customs.

Its limits, for an agent

Quotes and samples take back-and-forth. Each seller is a separate deal, with nothing linking one step to the next. Freight, customs and checking quality are yours to arrange.

3. Put custom work out to bid

The agent describes the work in plain English: what, how many, where it goes, by when, and the most the person would pay. Établi plans it into legs, one per kind of workshop the work needs; the agent drafts the project, and the person claims and publishes it, after which independent workshops bid on each leg. The person chooses offers and gets a contract per leg, and each order after that is one API call.

Use it when

• The work is custom: your design, your terms.
• It takes more than one workshop, such as pots made overseas for a nursery in the US.
• It goes to many addresses (a parcel to each person, an order per customer) or repeats.

Its limits

• No instant price: a preview has none, and prices come from bids.
• Orders go to US addresses only.
• When no workshop on Établi fits a leg yet, finding one can take a few days.
• No account needed to start: the agent drafts, and the person claims and publishes on etabli.io. Online payment isn't live yet: Établi arranges payment once the person chooses offers.

Side by side

When Établi is the wrong choice

• You need it today, or a price before you commit. Buy from a catalog.
• It's a standard product at high volume and you have someone to source it. A marketplace gives that person more control.
• It ships outside the US. Établi's orders go to US addresses.
• You want a long track record. Établi went live in October 2026; there are no public reviews of it yet.

Use it from your agent

Add Établi's MCP server (https://api.etabli.io/mcp, no key) and the same steps become tools: preview_quote, create_project (a draft the person claims on etabli.io) and get_project; with a buyer key, awarding and ordering. Or install the skill, which tells an agent when Établi fits.

• Claude Code

  claude mcp add --transport http etabli https://api.etabli.io/mcp

• Codex

  codex mcp add etabli --url https://api.etabli.io/mcp

  Or add it to ~/.codex/config.toml, which the Codex CLI, the IDE extension and the ChatGPT desktop app share:

  ~/.codex/config.toml

  [mcp_servers.etabli]url = "https://api.etabli.io/mcp"

• Cursor

  Add to Cursor

  Or open cursor.com/install-mcp in a browser.

• VS Code (Copilot)

  Add to VS Code

  Or run code --add-mcp '{"name":"etabli","type":"http","url":"https://api.etabli.io/mcp"}'.

• Gemini CLI

  gemini mcp add --transport http etabli https://api.etabli.io/mcp

• Skill (70+ agents)

  npx skills add https://etabli.io

  Installs Établi's skill, order-physical-work, into the agents on your machine: when to use Établi and how to order with it.

Every other client, and the tools: the MCP server. The API: docs.

More guides

• Dropship houseplants in your own pots
• Printed packaging, split across warehouses
• Get products manufactured
• An API for AI agents

The API, auth and every endpoint: docs. For agents, everything in one file: llms-full.txt.