Pepesto Mcp

Pepesto MCP Server

MCP server for the Pepesto API — give your agent the ability to turn any recipe (a URL, plain text, or a photo) into a matched basket of real supermarket products with live prices, across 26 European supermarkets. The MCP covers the recipe → matched cart half of the workflow (parse / search / map ingredients to SKUs / check catalogs); placing the actual order is a separate step — see Where checkout actually happens.

Quick install

Claude Desktop

Add to claude_desktop_config.json:

{  "mcpServers": {    "pepesto": {      "command": "npx",      "args": ["-y", "@pepesto/pepesto-mcp"],      "env": { "PEPESTO_API_KEY": "pep_sk_…" }    }  }}

Claude Code

claude mcp add pepesto -e PEPESTO_API_KEY=pep_sk_… -- npx -y @pepesto/pepesto-mcp

Getting an API key

Most tools need a key, but pepesto_predirect is public and free — it works with no key at all (the end user pays when they check out in the app).

1. Start with a pay-as-you-go credit pack — see https://www.pepesto.com/pricing/.

2. Mint an API key by calling /link with the email you used at checkout. The key is returned only once — store it immediately.

   bash

   curl -X POST https://s.pepesto.com/api/link \  -H "Content-Type: application/json" \  -d '{"email":"[email protected]"}'

3. Set the key in your environment:

   bash

   export PEPESTO_API_KEY=pep_sk_…

Tools

The MCP stops at "matched cart with prices" — see Where checkout actually happens for how users finish the order. /session, /checkout, and /link are intentionally not wrapped; see Roadmap for what's planned.

Example conversations

Quick: recipe URL → matched cart

The fastest path. One tool call returns a matched cart and a checkout link.

User: Use the BBC Good Food pizza margherita recipe to build a Tesco cart, and also add sparkling water and olive oil.

Assistant: [Uses pepesto_oneshot with content_urls, content_text, supermarket_domain: "tesco.com"]

Assistant: Cart matched at Tesco. Checkout link: <redirect_url>.

Free, deferred handoff → end user pays (pepesto_predirect)

When the API client doesn't want to pay for matching and a deferred deep link is acceptable. pepesto_predirect is free and needs no API key: it returns instantly with a link, and parsing + product matching happen lazily once the user opens it. The user is charged when they proceed to checkout in the Pepesto app (if the app isn't installed, they're sent to the app store first and the list is preserved).

Who this is for: pepesto_predirect is an end-user / agent-facing tool, not a developer-integration endpoint. It's the right fit when a person is chatting with an assistant (Claude Desktop, ChatGPT, etc.) and wants to hand their shopping list off to their own phone — the agent returns a link, the user finishes (and pays) in the app. If you're a developer building a product on top of Pepesto, you'll usually want pepesto_oneshot (matched cart up front, you pay) instead. Got a developer use case for predirect anyway? Tell us — we'd love to hear it.

Use pepesto_predirect when the cost should fall on the end user; use pepesto_oneshot when the client wants the basket matched up front (with prices) and is willing to pay for it.

User: Here's my weekly shopping list — just give me a link I can open on my phone to finish in the Pepesto app.

Assistant: [Uses pepesto_predirect with shopping_list: "2 avocados\n1 loaf of bread\n500 g tomatoes\n..."]

Assistant: Done — 🛒 Open your shopping list in Pepesto → Free to open · 3 items ready · you only pay at checkout in the app. Open it on your phone (on a computer it'll show a QR code to scan).

The tool returns this as ready-to-show Markdown — one tappable, labeled link plus a short caption — so the long redirect_url is never dumped as raw text.

Browse the recipe graph → pick → matched cart

For "find me something to cook" prompts, where the agent searches Pepesto's recipe graph first and returns matched products with prices.

User: Find me a vegan pasta recipe for two.

Assistant: [Uses pepesto_suggest with query: "vegan pasta dinner for two"]

Assistant: Three options: Creamy Cashew Penne, Lemon Garlic Spaghetti, Roasted Veg Rigatoni. Which would you like?

User: Go with the first one. Shop it at Coop (coop.ch).

Assistant: [Uses pepesto_products with the chosen recipe's kg_token and supermarket_domain: "coop.ch"]

Assistant: Matched 8 ingredients at Coop — cashews CHF 4.20, penne CHF 1.95, ... — total CHF 18.40.

Parse a specific recipe → matched cart

When the user already has a recipe URL or text in mind. The MCP returns the matched product list — what to do with it is up to you.

User: Parse this and break out the ingredients: https://www.bbcgoodfood.com/recipes/pizza-margherita-4-easy-steps

Assistant: [Uses pepesto_parse with recipe_url]

Assistant: Pizza Margherita in 4 easy steps — 8 ingredients: strong bread flour, fast-action yeast, ...

User: Now match those to Sainsbury's products.

Assistant: [Uses pepesto_products with the returned kg_token and supermarket_domain: "sainsburys.co.uk"]

Assistant: Matched 8 ingredients at Sainsbury's — total £6.40.

Weekly meal plan → one merged matched cart

Multiple kg_tokens through pepesto_products in a single call — Pepesto merges duplicated ingredients across recipes.

User: Plan 5 keto dinners for the week and match them at Delhaize.

Assistant: [Uses pepesto_suggest with query: "keto dinner low carb high protein", num_to_fetch: 5]

Assistant: Five dinners: Steak & Kale, Salmon Avocado Bowl, Zucchini Lasagna, Cauliflower Risotto, Chicken Caesar.

User: Build one shopping list that covers the whole week.

Assistant: [Uses pepesto_products with all five kg_tokens and supermarket_domain: "delhaize.be"]

Assistant: Combined match: 24 unique ingredients (some shared across recipes), estimated total €68.40 at Delhaize.

Compare prices across two supermarkets

Run the same recipe through pepesto_products for each market and compare totals.

User: Parse this Sunday lunch recipe.

Assistant: [Uses pepesto_parse]

Assistant: Got it: Italian Sunday Lunch, 12 ingredients.

User: Show me the cart at both Esselunga and Conad so I can compare.

Assistant: [Uses pepesto_products twice — supermarket_domain: "spesaonline.esselunga.it", then supermarket_domain: "spesaonline.conad.it"]

Assistant: Esselunga: €34.20. Conad: €31.50. Conad is €2.70 cheaper, mostly on pasta and tomatoes.

Catalog dump for market analysis

Only on explicit request, and cache aggressively — see Pricing & best practices below.

User: I'm building a price-comparison dashboard. Pull the full Plus NL catalog.

Assistant: Heads up — pepesto_catalog is the most expensive call here. I'll only run it once; you should cache the result for at least a day. Confirm?

User: Yes.

Assistant: [Uses pepesto_catalog with supermarket_domain: "plus.nl"]

Assistant: Catalog dumped: 1,847 SKUs across 23 categories.

Supported supermarkets

Need a supermarket that isn't on this list? Contact Pepesto.

Where checkout actually happens

This MCP stops at "matched cart with prices." It does not automate placing the order on the supermarket's website. Two ways to finish the trip:

• Pepesto app (recommended). Open the redirect_url returned by pepesto_oneshot in a browser, or hand the user the matched-product list from pepesto_products and tell them to recreate it in the Pepesto app — that's where the hosted checkout flow lives, including login, basket review, and (for some markets) payment.
• The supermarket's own site. The user can take the matched product list from pepesto_products and add the SKUs directly on tesco.com / coop.ch / etc. Slower, but no Pepesto account needed.

Pricing & best practices

Pepesto runs on simple pay-as-you-go credits — you only pay for what your agents actually use, and credits never expire, so a top-up is yours until you spend it. We also offer discounts for students and early-stage teams, so say hi if that sounds like you. Full per-call pricing and volume tiers live at https://www.pepesto.com/pricing/.

A few tips to get the most out of every credit:

• pepesto_credits is free — call it any time for a quick balance read-out.
• pepesto_predirect is free and needs no API key — it defers matching and bills the end user at checkout, so it costs the API client nothing.
• pepesto_oneshot, pepesto_parse, pepesto_suggest, and pepesto_products are the everyday calls (match a recipe, plan a week, compare baskets) and are priced for routine agent use.
• pepesto_catalog does a full SKU dump for a supermarket and is the heaviest call. It's the right tool for genuine market analysis or price-comparison dashboards — just cache the result for at least a day per supermarket. Not sure you need it? Tell us about your use case and we'll usually point you to a cheaper path.

Roadmap

Planned to follow:

• pepesto_session — wrap /session so an agent can build a Pepesto-side checkout session from selected SKUs.
• pepesto_checkout — wrap /checkout, the turn-by-turn browser-automation loop that drives the supermarket's own site (login, add-to-basket, prompt-for-CAPTCHA, etc.). This is the missing piece for fully autonomous shopping.
• Hosted-checkout handoff — surface the Pepesto-app deep link as a structured tool result (instead of free text), so MCP clients can render it as a button instead of a URL.

If any of these would unblock you, tell us — it'll move them up the queue.

Development

git clone https://github.com/pepesto-solutions/pepesto-mcp.gitcd pepesto-mcpnpm installnpm run buildnpm testnpm run test:coverage

Run the inspector against the local build:

PEPESTO_API_KEY=pep_sk_… npm run inspector

License

The Pepesto MCP server in this repository is licensed under the MIT License.

Deploy

• Deploy the Pepesto MCP on Glama.