ShopSavvy MCP¶
Two retailers can sell the same barcode at prices a hundred dollars apart, and no single storefront will tell you that. Price comparison is only useful when it is attached to a stable product identity: a barcode, an ASIN, a model number. ShopSavvy's catalogue is built on exactly that identity, so a question like what does this cost anywhere becomes answerable from one identifier instead of from a search across tabs.
The server is keyless for reads, enumerates fifteen tools with read-only annotations, and answers in the assistant's own text rather than in a table the client must render. Three calls were driven this pass and the answers are checkable line by line.
Server type: remote Streamable HTTP, stateless (no Mcp-Session-Id issued)
Endpoint: https://api.shopsavvy.com/mcp
Server: shopsavvy, version 2.0.0, title "ShopSavvy"
Protocol: 2025-06-18
Auth: none for reads; a key is required for scheduled re-checks and usage reporting
Tools: 15 (annotated read-only or account-scoped)
Repository: github.com/shopsavvy/shopsavvy-mcp-server
Listed on: mcp.so submission queue #7496
The evidence tier¶
Keyless and driven. initialize answered 200 as shopsavvy 2.0.0 on protocol 2025-06-18 with no session identifier and no credential, and tools/list returned fifteen tools, most carrying readOnlyHint: true. Three were called:
product_searchon instant pot duo 6 quart answered 20 of 72 total matches with brand, category path, barcode, ASIN and a relevance score on each row, which is how a product identity is obtained in the first place.product_offerson barcode 853084004088 answered 25 offers checked in the last 90 days, lowest first, from $38.22 at Truegether (used) through $44.99 at BUYA.COM to $45.00 at two storefronts, each with availability, condition, seller and a last-checked timestamp.product_price_historyfor 2026-11-01 to 2026-11-16 answered a per-retailer series: Walmart at $142.98 and $143.10 on 11/15, and Crate & Barrel at $89.99 on the same day for the same barcode.
That spread is the product. A buyer asking whether $142 is a good price gets an answer with a second retailer named, and a seller asking where they sit gets the same list from the other side.
Two defects are recorded rather than smoothed. The history answer lists several dozen retailers whose rows read No historical data available, including a handful of storefront names that look like catalogue noise rather than shops, and Walmart appears twice with two prices a few cents apart. A reader who takes the retailer count as evidence of coverage will overread it; the price points themselves are the part that reconciles.
Tools & Capabilities¶
| Tool | Required arguments | What it answers |
|---|---|---|
product_lookup |
identifier |
One product record: title, brand, model, MPN, barcode, ASIN, ShopSavvy ID |
product_lookup_batch |
identifiers |
The same, for a comma-separated list of barcodes, ASINs, URLs or model numbers |
product_search |
query |
Ranked catalogue matches with brand, category path and identifiers |
product_offers |
identifier |
Current prices at every retailer with an offer, lowest first |
product_offers_retailer |
identifier, retailer |
The same, at one domain such as amazon.com or bestbuy.com |
product_price_history |
identifier, start_date, end_date |
Recorded prices per retailer over a window that ends today or earlier |
product_schedule |
identifiers, schedule |
Hourly, daily or weekly automatic re-checks on the key's account |
product_unschedule |
identifiers |
Stops those re-checks |
scheduled_products_list |
none | The watchlist, grouped by frequency |
api_usage |
none | Credits used, monthly limit, remaining, reset date |
content_best_picks |
query |
Published best-of buying guides for a category |
content_tldr_review |
none | A short review summary with a score out of ten |
content_qa_answers |
identifier |
Published answers to common questions about a product |
content_buying_guide |
query |
When-to-buy guidance by category or brand |
universal_search |
query |
Category search with pre-computed rankings |
Reads are keyless. Scheduling, the watchlist and credit accounting sit behind an API key, which is the same shape the vendor's own MCP repository documents.
Installation¶
claude mcp add shopsavvy --transport http https://api.shopsavvy.com/mcp
The submitter's listing names an SSE route at /mcp/sse; that path answers 404 to a POST, while /mcp answers the JSON-RPC handshake directly. Use the bare /mcp address.
Why This Matters for Operators¶
Retail pricing decisions fail in two directions: buying at a price that another retailer undercuts, and pricing a product above a competitor the seller never checked. Both need the same primitive, a barcode resolving to every current offer plus its history. product_schedule turns that from a lookup into a watchlist, and product_lookup_batch makes a purchase order or a competitor set checkable in one call rather than one call per item.
Questions operators ask¶
Is the pricing data live or cached?¶
Cached with timestamps. Each offer carries its own last checked time, and the offers call reports how many were checked in the last ninety days, so an agent can say how stale an answer is instead of implying it is current.
Can it watch prices over time?¶
Yes, on a key: product_schedule sets hourly, daily or weekly re-checks and scheduled_products_list reads the watchlist back. Without a key the history tool still answers, from ShopSavvy's recorded points rather than from a schedule you own.
What is the coverage limit?¶
The search answered 72 matches for a common kitchen appliance, which is a curated catalogue rather than the whole web. Niche or private-label items may not resolve, and product_lookup is the honest test before any workflow depends on one.
Related¶
- SpendCanary MCP - the advertising side of the same margin question
- Vizbl MCP - the product-page side of ecommerce conversion
- Shopify Speed Check MCP - storefront performance for the same catalogue
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.