roastify-mcp
MCP server for the Roastify Developer API, letting AI agents browse the coffee catalog and generate packaging artwork, with paid tools settled in Bitcoin Lightning.
README
roastify-mcp
A Tollbooth DPYC operator for the Roastify Developer API. Roastify is a white-label coffee dropshipper: it roasts, prints branded packaging, and ships direct to customers.
This operator lets an AI agent read the Roastify catalog and generate packaging artwork, paid for in Bitcoin Lightning satoshis with no account, no KYC, and no subscription.
What you're building
YOU (operator, human in the loop)
│ │
│ set & tune prices │ drive credential intake
▼ ▼
┌───────────────┐ ┌──────────────────────────────────────────────┐
│ Pricing Studio│ prices │ Roastify — YOUR OPERATOR MCP │
│ (iOS) ├───────▶│ FastMCP · deployed on Horizon │
└───────────────┘ Neon │ ┌────────────────────────────────────────┐ │
│ │ roastify domain module ◀─ Roastify API │ │
Patron (Citizen) │ │ @runtime.paid_tool(FROZEN_UUID) tools │ │
+ MCP client ─────────▶│ ├────────────────────────────────────────┤ │
(Claude, etc.) npub │ │ tollbooth-dpyc SDK (the wheel) │ │
+sats │ │ ledger · vault (AES-256-GCM) · pricing │ │
│ │ /ConstraintGate · Secure Courier·audit │ │
│ └────────────────────────────────────────┘ │
└───┬─────────┬──────────┬───────────┬─────────┘
▼ ▼ ▼ ▼
Neon Postgres BTCPay▶ Sponsor Nostr relays
(your schema) Lightning Authority proofs·courier
ledger+pricing invoices certify + DMs·audit
provision │
api.roastify.app ◀── domain │ ▼
(each patron's calls └──▶ DPYC Oracle +
own API key) dpyc-community
Onboarding roadmap
- Nostr keypair — generate one (
nak key generate). The nsec is the only env var:TOLLBOOTH_NOSTR_OPERATOR_NSEC. - Sponsor Authority — register; it provisions your Neon schema automatically.
- Secure Courier — deliver your BTCPay secrets (
btcpay_host,btcpay_api_key,btcpay_store_id) by Nostr DM. They are vaulted, never put in the environment. - Set prices in Pricing Studio — see below. Tools start unpriced and uncallable.
- Deploy on Prefect Horizon —
.fastmcp.yamlis ready.
Get Pricing Studio
Your tools ship unpriced, and nobody can call a paid tool that has no price. Pricing Studio (iOS) is the operator console: it reads and writes the pricing model live in Neon, so prices never live in code — see tollbooth-pricing-studio.
That indirection is the point. Dynamic per-tool pricing — surge, happy-hour, loyalty discounts, free trials for a first-time patron — is what DPYC gives you that a flat paywall never could.
Bring your own Roastify key
Every patron delivers their own Roastify API key through the Secure Courier, vaulted per npub. There is no operator-held key and no fallback path.
This follows from the domain rather than being a policy choice. Roastify scopes the catalog, saved designs, and plan tier to the merchant account behind the key — a patron on Base sees a different catalog than one on Pro. A single operator-held key would hand one merchant's world to every caller, which is a wrong answer rather than a limitation to work around. It also means this operator never holds a credential it could misuse, and revoking your key in the Merchant App severs access without involving the operator.
roastify_get_patron_onboarding_status # what's missing
roastify_request_patron_credentials # → you get a Nostr DM
(reply with your Roastify API key)
roastify_receive_patron_credentials # vaulted, encrypted at rest
Live keys start rty_live_. Sandbox keys start rty_test_ and do not fulfill — useful for
exercising the read tools without touching production.
Tools
| Tool | Tier | What it does |
|---|---|---|
roastify_browse_catalog |
read | Catalog products and coffee blends together, plan tier marked |
roastify_get_catalog_product |
read | One product with variants, prices, stock, dieline template |
roastify_get_blend |
read | One blend: roast level, decaf status, variants |
roastify_list_my_products |
read | Your saved designs, cursor-paginated |
roastify_get_my_product |
read | One saved design in full |
roastify_check_stock |
read | Stock for one SKU or the whole list |
roastify_generate_artwork |
write | Generate artwork from a saved design |
roastify_artwork_status |
free | Check that job |
Plus the standard DPYC catalog (balance, purchase, courier, pricing, Oracle, status).
What this operator does not do, and why
It cannot create, update, delete, or sync products. Not a missing feature — the Roastify
API has no such endpoints. /v1/products is read-only, and the words integration, sync,
shopify, and upload do not appear anywhere in the
OpenAPI spec. Saved products and storefront sync are
Merchant App capabilities by design.
It does not place orders. Roastify already routes orders to its own operations and reports tracking back to your storefront. An MCP layer between those adds a hop without adding judgment, so order automation is deliberately left to Shopify and Roastify.
Artwork generation is template personalization, not design. roastify_generate_artwork
rewrites the named text and image placeholders of a design you already authored in the
Roastify Design Studio. It cannot author a design from scratch.
A generated artwork URL is not attached to anything. In Roastify's API the only consumer of an artwork URL is an order line — and orders are out of scope here. You receive a URL and carry it onward yourself. That seam is accepted deliberately; it is not an oversight, and this operator will not simulate a design flow the API cannot complete.
The catalog has no palate data. Origin, region, altitude, processing method, and varietal
exist only in the Merchant App UI. The machine-readable signals are roastLevel, isDecaf,
size, and price; everything else about a coffee's character is prose in its description. Bean
selection is correspondingly weaker than it looks, and this operator does not dress up an
LLM's reading of a description as structured provenance data.
Artwork is asynchronous
Roastify renders artwork on its own servers, so roastify_generate_artwork hands back a job
id straight away and roastify_artwork_status checks it. Status checks are free — you
learn the work finished by looking, and charging for each look would be charging for waiting.
job = roastify_generate_artwork(product_id=..., fields=[...], client_req_id="my-req-1")
roastify_artwork_status(job_id=job["job_id"]) # free, poll until artwork_url appears
Pass client_req_id and a repeated request is safe: it becomes Roastify's
Idempotency-Key, so the artwork is not generated twice.
The operator does no waiting of its own — both calls return immediately — so there is no detached executor here and nothing to recycle. That machinery would earn its place only if this operator ever rendered artwork itself.
Development
pip install -e ".[dev]"
ruff check .
pytest -v
python -m roastify_mcp.server # needs TOLLBOOTH_NOSTR_OPERATOR_NSEC to fully run
Upstream status
Roastify's API is version 0.3.1 and its own spec marks it beta, "subject to change." Expect to follow breaking upstream edits, and weigh that before promising this operator's catalog to third-party merchants.
License
Apache 2.0 — see LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。