galaxus-mcp
Product search and discovery on Galaxus and Digitec online shops: search, browse, compare, look up prices and specs.
Click on "Deploy Server".
Wait a few minutes for the server to deploy. Once ready, it will show a "Started" state.
In the chat, type
@followed by the MCP server name and your instructions, e.g., "@galaxus-mcp Find a wireless mouse under CHF 50 with good ratings"
That's it! The server will respond to your query, and you can continue using it as needed.
Here is a step-by-step guide with screenshots.
galaxus-mcp
An MCP server for product search and discovery on Galaxus and Digitec. Read-only: search, browse, compare, look up prices and specs. Nothing that requires a login (no cart, no orders, no account data).
Tools
Tool | What it does |
| Full-text search. Returns products plus the facets you can narrow with. |
| Everything about one product: price, stock, specs, variants, price history, warranty. |
| Type-ahead: search-term suggestions and a few direct product hits. |
| List a category (product type) without a search term. |
| List a brand's products. |
| Alternatives ( |
| Average rating, rating count, and the pros/cons reviewers mention most. |
Every list tool supports filters, min_price / max_price, sort
(RELEVANCE, LOWEST_PRICE, HIGHEST_PRICE, RATING, NEWEST, AVAILABILITY), limit and
cursor pagination.
Filters are discoverable rather than hardcoded: search and browse results carry a filters block
listing each facet's id and its options with counts (bra = brand, pt = category, pr = price
range, rating, plus per-category spec filters such as "Signal transmission"). Feed those ids
straight back in:
{ "query": "wireless mouse", "filters": [{ "filter_id": "bra", "option_ids": ["292"] }], "max_price": 80 }Related MCP server: galaxus-mcp
Setup
npm install
npm run build
npm run smoke # exercises all 7 tools against the live shopThen register it with Claude Code:
claude mcp add galaxus -- node /absolute/path/to/galaxus-mcp/dist/index.jsOr drop it in an MCP client config (this repo also ships a project-scoped .mcp.json):
{
"mcpServers": {
"galaxus": {
"command": "node",
"args": ["/absolute/path/to/galaxus-mcp/dist/index.js"],
"env": { "GALAXUS_PORTAL": "galaxus", "GALAXUS_LANGUAGE": "en" }
}
}
}Env var | Values | Default |
|
|
|
|
|
|
The language affects product names, specification labels and facet titles.
Using it
Once registered, just ask in natural language — the tools chain on their own:
"Find me a wireless mouse under CHF 100 with good reviews" →
galaxus_searchwithmax_priceandsort: RATING, thengalaxus_get_review_summaryon the pick."What are the specs of the Logitech MX Master 3S, and has it been cheaper?" →
galaxus_get_product, which carries the specs and the price-history summary."Show me alternatives to this one that are cheaper" →
galaxus_related_productswithkind: similar."What Logitech keyboards are in stock?" →
galaxus_browse_brandwithsort: AVAILABILITY.
Product ids are the trailing number in any Galaxus URL, and the product tools accept the full URL too, so pasting a link works.
How it talks to the shop
The storefront exposes a GraphQL API that only accepts persisted operations: the operation's
hash is part of the URL (/graphql/o/<hash>/<operationName>) and the request body carries variables
only. Query text is rejected with a 404, so there is no schema introspection and no arbitrary
queries — this server replays the same operations the website itself uses.
The shop is also behind bot protection that rejects plain curl and headless Chromium alike. The
persisted endpoints, however, answer ordinary fetch calls, so the running server needs no
browser — a browser is only involved when refreshing hashes, below.
Hash rotation: the one thing that will break this
The operation hashes live in src/operations.json. They are tied to the deployed frontend build, so
Galaxus rotates them whenever it ships a new frontend — which happens often. When that occurs,
the URLs this server calls no longer exist and every tool starts failing at once.
Symptom. Every tool call comes back with:
The persisted hash for "useSearchDataQuery" is no longer accepted by the shop. Galaxus deployed a new frontend and rotated its query hashes. Run
npm run refresh-hashes(thennpm run build) to re-capture them.
Note the shop reports this in-band: HTTP 200 with the GraphQL error
The specified persisted operation key is invalid. — not a 404. So it cannot be mistaken for a
network problem.
Checking, without a browser. Because the shop validates the hash before the variables, posting an operation with empty variables tells you whether its hash is alive: a live one complains about a missing variable, a rotated one rejects the key. That is one tiny request per operation:
npm run check-hashes # exits 0 if all hashes are current, 1 if any rotatedFixing. Re-capture and rebuild:
npx playwright install chromium # once, if you have not already
npm run refresh-hashes # opens a real browser window — let it finish
npm run build
npm run smoke # confirm all 7 tools are green againrefresh-hashes drives a real browser across a search page, a category page, a product page and a
brand page. It harvests hashes two ways, because neither alone catches everything: from the Relay
artifacts embedded in the JS bundles (params:{id:"<hash>",…,name:"<operation>"}) and from the
GraphQL requests those pages actually fire. It then rewrites src/operations.json in place, logging
every hash that moved, and keeps the previous value (exiting non-zero) for any operation it did not
see, rather than writing a broken one.
It runs headed on purpose — headless Chromium gets blocked. Expect a browser window for about a
minute; don't close it. Commit the resulting src/operations.json diff.
Automatically. .github/workflows/refresh-hashes.yml runs check-hashes daily. That step needs
no browser, so the usual run is cheap and silent. Only when a hash has actually rotated does it
install Chromium, re-capture the hashes with a headed browser on a virtual display
(xvfb-run, since headless is blocked), re-verify with check-hashes and smoke, and open a PR
with the new src/operations.json. You can also trigger it by hand from the Actions tab, with
force to re-capture even while the current hashes still work.
One caveat: the workflow talks to Galaxus from a GitHub-hosted runner, and the shop's bot protection
judges by IP as well as by browser. If those datacenter IPs turn out to be blocked, the capture step
will fail there — run npm run refresh-hashes locally instead (or point the workflow at a
self-hosted runner). Nothing else about the server depends on this: it is a maintenance path only.
Two behaviours worth knowing
Search redirects. A generic query like mouse makes the shop return a category redirect instead
of products. The server always sends skipRedirect, so you get products back.
The price filter is family-wide. The shop matches a product when any of its variants falls in
the price range, while the price it shows is the cheapest variant — so a CHF 100–200 filter would
otherwise surface an CHF 82.90 mouse that happens to have a CHF 3860 colourway. The list tools
therefore drop products whose own price falls outside min_price/max_price, and page forward to
refill the page. Pass strict_price: false to see the shop's raw behaviour.
Scope
Search and discovery only. Anything behind a login — cart, checkout, orders, wishlists, writing
reviews — is deliberately out of scope. total_results reflects the shop's own count, which counts
product families, so it can exceed the number of rows returned once strict price filtering applies.
Individual review texts are not exposed: the shop renders them server-side and no persisted
operation returns them, so galaxus_get_review_summary gives the rating summary and the aggregated
pros/cons keywords instead.
Available Tools
7 toolsgalaxus_autocompleteAutocomplete a search termARead-only
Fast type-ahead lookup. Returns search-term suggestions (with the categories they map to) and a handful of directly matching products. Useful to resolve a vague phrase into a concrete product or category before searching.
Inputs: query (required).
galaxus_browse_brandBrowse a brandARead-only
List products from one brand. Brand ids come from the bra facet returned by galaxus_search or galaxus_browse_category.
Inputs: brand_id (required), sort, limit, cursor, +4 more.
galaxus_browse_categoryBrowse a categoryARead-only
List products in a category (product type) without a search term — the equivalent of opening a category page. Category ids come from galaxus_search results (category_id), the pt facet, or galaxus_autocomplete.
Inputs: category_id (required), sort, limit, cursor, +4 more.
galaxus_get_productGet product detailsARead-only
Full detail for one product: price and previous price, availability and stock, rating, full specifications, variants, price-history summary, warranty and return policy. Accepts a numeric product id or a Galaxus product URL.
Inputs: product (required).
galaxus_get_review_summarySummarise what reviewers sayARead-only
What customers think of a product: the average rating, how many people rated it, and the pros and cons reviewers mention most often. Individual review texts are not exposed by the shop's API.
Inputs: product (required).
galaxus_searchSearch productsARead-only
Full-text product search across the shop. Returns matching products with prices, ratings and availability, plus the facets (brand, category, price range, specs) that can be fed back in as filters to narrow the search.
Inputs: query (required), sort, limit, cursor, +4 more.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
TDQS
Scored across 7 tools
Each tool has a clearly distinct purpose: autocomplete for suggestions, search for query-based results, get_product for details, browse_category and browse_brand for filtered browsing, related_products for discovery, and review_summary for ratings. There is no overlap; even browse_category and browse_brand are distinguished by their governing attribute (category vs. brand).
All tools share the 'galaxus_' prefix, which aids recognition. However, after the prefix the pattern varies: 'search' and 'autocomplete' are bare verbs, 'get_product' and 'get_review_summary' use get_noun, 'browse_category' and 'browse_brand' use verb_noun, while 'related_products' is a noun phrase without a verb. This is a minor deviation that is still predictable and readable.
Seven tools is an ideal size for a product information server—small enough to be easily navigable, yet comprehensive enough to cover search, browsing, retrieval, and related recommendations. Each tool fills a distinct niche without redundancy.
The tool set covers the core workflow of product discovery: autocomplete, search, browse by category/brand, full details, related items, and review summaries. A minor gap is that you cannot enumerate all categories or brands independently without first performing a search, but facet-based access is a reasonable design. No dead ends appear for typical shopping research.
Maintenance
Related MCP Connectors
Compare prices across Swiss and European shops — barcode (GTIN) lookup and daily price history.
Search products, compare prices and discover deals across 6 European markets with your AI assistant.
Search and look up products across thousands of online stores, normalized into one schema.
Search and look up products across thousands of online stores, normalized into one schema.
Related MCP Servers
- AlicenseAqualityDmaintenanceEnables searching geizhals.de for products and retrieving shop prices and offers to assist with price comparison.2MIT
- AlicenseAqualityCmaintenanceEnables AI assistants to search products and retrieve detailed information from Galaxus and Digitec, including prices, specifications, and price history.54MIT
- AlicenseNot gradedqualityBmaintenanceEnables AI to search Geizhals price-comparison data, including products, best prices, price history, ratings, categories, and deals.8MIT
- AlicenseAqualityCmaintenanceEnables live product search, price, stock, and spec queries across Flipkart and Amazon.in, with tools for comparing products and checking delivery availability.123MIT