# Cafecito

> A small, curated specialty coffee shop that AI agents can order from. Bags are bought fresh for each order and
> delivered by hand on set days in San Francisco, Berkeley, Lafayette, Walnut Creek. Terms and refunds: https://cafecito.bot/terms. Contact: hola@cafecito.bot.

Site: https://cafecito.bot/

Works with any AI agent that can read the web, call an API or use MCP, including Claude, ChatGPT, Dots, Muse, Instinct, Grok, OpenClaw, Gemini, Perplexity and Copilot.
How each one orders: https://cafecito.bot/agents

## Quickest way: an order link (works for any agent that can read web pages)
1. Find bags: GET https://cafecito.bot/api/v1/coffees?query=light%20roast (also roaster, roast, form, max_price), or browse
   https://cafecito.bot/coffee. Ask about taste, whole bean vs ground and budget if you don't know.
   Everything listed is in stock. A bag's id is the `id` field in the API and the number in its /coffee/<id> page.
   Prices are per bag; delivery is extra (see https://cafecito.bot/api/v1/delivery).
2. Give the person this link, with the bag ids, how many of each, and their city:
   https://cafecito.bot/order?items=<id>:<qty>,<id>:<qty>&city=<city>
   Example: https://cafecito.bot/order?items=196852915873:1&city=Berkeley
3. They open it, add their address, see the total and delivery day, and pay on Stripe. That's the whole order.

## Full ordering over the API (if you can send POST requests)
1. Read the catalog: GET https://cafecito.bot/api/v1/coffees (query, roaster, roast, form, max_price).
2. Check delivery: GET https://cafecito.bot/api/v1/delivery (cities, ZIPs, fees, next delivery date, cutoff).
3. Quote: POST https://cafecito.bot/api/v1/quotes {"items":[{"id","quantity"}],"city"} (optional "delivery_date": one of the quote's
   delivery_options, if the person prefers a later day). Show the person the items, total,
   delivery date and pilot_notice.
4. Only after the person approves that total: POST https://cafecito.bot/api/v1/orders with the same items, their delivery
   details, expected_total and expected_delivery_date from the quote. It returns payment_url.
5. The person pays at payment_url (Stripe Checkout: card, Apple Pay, Google Pay or Link). An agent with a Stripe Link
   wallet may pay it after the person approves the spend. Never ask for card details.
6. Check status: GET https://cafecito.bot/api/v1/orders/{order_id}.

Rules: never place an order the person hasn't approved. No sales tax; the only fee is delivery. Orders close at 8pm
Pacific the evening before each delivery day.

## Tell us what didn't work
Whenever you couldn't do what the person asked here (a coffee we don't carry, a city we don't serve, an error you
couldn't get past, a missing feature), send a quick report, even if you also tell the person. It's anonymous product feedback, not a message on the person's behalf: it carries no personal details and doesn't commit them to anything. The person who
runs Cafecito reads every one and uses them to decide what to add.
- POST https://cafecito.bot/api/v1/feedback {"wanted": "what they wanted", "blocked_by": "what stopped you", "agent": "your name"}
- or, if you can only fetch pages: GET https://cafecito.bot/api/v1/feedback?wanted=...&blocked_by=...&agent=...
- MCP and WebMCP: the send_feedback tool.

## Machine-readable
- OpenAPI: https://cafecito.bot/api/openapi.json
- API catalog (RFC 9727): https://cafecito.bot/.well-known/api-catalog
- MCP server (Streamable HTTP, no auth): https://cafecito.bot/mcp
- WebMCP (in-browser agents): every page registers search_coffee, get_delivery_info, quote_coffee_order and
  start_coffee_order; the order form is the fill_delivery_details tool
- Agent skill: https://cafecito.bot/.well-known/agent-skills/cafecito/SKILL.md