Swissgroceries MCP

swissgroceries-mcp

Real-time Swiss grocery shopping over the Model Context Protocol. Search products, compare prices across Migros, Coop, Aldi, Denner, Lidl, Volgshop, Otto's, and Coop City, see weekly promotions, and plan multi-store shopping trips. Works with any MCP-compatible client (Claude Desktop, Claude Code, Cursor, Cline, Continue, VS Code MCP extensions, custom clients).

Disclaimer

This is a personal fun project. It is not affiliated with, endorsed by, or sponsored by Migros, Coop, Aldi, Denner, Lidl, Volg, Otto's, or any other retailer. It uses publicly accessible mobile-app endpoints to make Swiss grocery shopping a bit smarter for end users.

If you represent any of these stores and have concerns (about API usage, branding, scraping rate, or anything else), please reach out to the maintainer through GitHub and we will work it out. No need to escalate.

API stability: the chain APIs used here are unofficial and can change at any time. The maintainer is not responsible for failures caused by upstream changes; please open an issue with the response sample so the affected adapter can be updated.

PRs welcome. New chains, better matchers, smarter strategies, bug fixes, doc improvements; all encouraged. See CONTRIBUTING.md.

Install

No accounts, no tokens, no API keys required. The Denner adapter self-registers an anonymous client on first use; everything else uses public endpoints.

Claude Desktop (one-click)

Download swissgroceries-mcp.mcpb from the Releases page and:

• macOS: double-click or drag onto the Claude Desktop app icon.
• Windows: Settings → Extensions → Advanced → Install Extension → select the file.

Claude Code (one-liner)

claude mcp add swissgroceries -- npx -y @nicktcode/swissgroceries-mcp

Cursor / Cline / Continue / VS Code / Claude Desktop (manual config)

Most MCP-compatible clients accept the same JSON server entry. Add it to your client's MCP config file (paths vary, see your client's docs):

{

"mcpServers": { "swissgroceries": { "command": "npx", "args": ["-y", "@nicktcode/swissgroceries-mcp"] } } }

Common config paths:

• Claude Desktop: ~/Library/Application Support/Claude/claude_desktop_config.json (macOS), %APPDATA%\Claude\claude_desktop_config.json (Windows).
• Cursor: .cursor/mcp.json in the project, or ~/.cursor/mcp.json globally.
• Cline / Continue / VS Code: see each client's MCP documentation.
• Custom clients: any stdio-based MCP client can spawn npx -y @nicktcode/swissgroceries-mcp directly.

What you can ask

Price comparison

• "Where is milk cheapest near 8001 Zürich right now?"
• "Compare pasta prices across Migros and Coop."
• "Show me organic milk options under CHF 3."

Shopping planning

• "I need milk, bread, eggs, chicken, and pasta near 8050. Where should I shop to keep costs down?"
• "Plan my weekly shop for 5 items near 4052 Basel, one stop only."
• "Split my cart across stores for the absolute lowest total, but add a 2 CHF penalty per extra trip."

Promotions and deals

• "What is on sale at Aldi this week?"
• "Any Migros deals on cheese ending this week?"
• "Show me all promotions across chains for pasta."

Stores and stock

• "Find Coop stores within 3 km of Bern Hauptbahnhof."
• "Which Migros near me has product 4389992 in stock?"
• "List Denner branches near 8050."

Tools

Each tool exposes rich JSON Schema with field-level descriptions, so the LLM knows when and how to call it.

Prompts

Prompts are surfaced as one-click templates in MCP-aware clients (Claude Desktop, Cursor, Cline). They produce structured user messages that drive the right tool calls — useful as a starting point and as discovery of what the server can do.

Resources

Chain coverage

NormalizedProduct.nutrition is normalized to a per-100g (or per-100ml) basis with the standard Swiss labelling fields — energyKj, energyKcal, fat, saturatedFat, carbs, sugar, fiber, protein, salt — so consumers can sort across chains ("highest protein per 100g lasagne") even though each upstream API exposes the data on a different surface and in a different shape. Missing or unparseable fields are left undefined; adapters never fabricate values.

Configuration

How it works

Each grocery chain is wrapped in an independent adapter (src/adapters/ /) that handles authentication, HTTP calls, and raw-to-normalised mapping. Adapters all produce the same NormalizedProduct, NormalizedStore, and NormalizedPromotion shapes (defined in src/adapters/types.ts), so the rest of the system never has to know which chain it is talking to.

The HTTP utility (src/util/http.ts) underpins every adapter except Migros (which delegates to the migros-api-wrapper library): in-memory response caching with a 5-minute TTL, retry with exponential backoff (3 attempts, 250 ms base), per-host rate limiting (~10 requests per second), and a per-host circuit breaker that opens after 5 consecutive failures and resets after 60 seconds.

The shopping planner (src/services/planner.ts) fans out store and product searches in parallel across all active adapters, then feeds results into a strategy solver (src/services/strategy.ts) that supports three modes:

• single_store: minimise the number of stops.
• split_cart: cheapest split across chains, with a configurable per-stop penalty.
• absolute_cheapest: cheapest split, ignoring stop count.

Cross-chain comparisons are kept fair by a category-text canonicality filter (src/services/matcher.ts, isCanonical). When at least one chain returns a product whose category text matches the query, results from chains that only returned tangential products (for example, Apfelschorle when searching for "apfel") are dropped from the comparison matrix for that item.

MCP client (any LLM)

│ │ MCP tool call ▼ src/index.ts ── buildRegistry() ────────────────────────────────────────┐ │ │ │ routes to tool handler │ ▼ ▼ src/tools/ src/adapters/ find_stores.ts ──► geocoding ──► adapter.searchStores migros/ search_products.ts ──────────────► adapter.searchProducts coop/ get_product.ts ──────────────► adapter.getProduct aldi/ get_promotions.ts ──────────────► adapter.getPromotions denner/ (auto-auth) find_stock.ts ──────────────► adapter.findStoresWithStock lidl/ plan_shopping.ts ──► geocoding ──► planner ──► strategy solver │ NormalizedProduct NormalizedStore NormalizedPromotion

Build from source

node --version   # requires Node.js >=20

git clone https://github.com/nicktcode/swissgroceries-mcp cd swissgroceries-mcp npm install npm run build

To also build the .mcpb bundle locally:

npx tsx scripts/build-mcpb.ts

Troubleshooting

**Coop