Newegg Refurbished / Recertified Memory Search

Important Background (Read First)

Newegg has no real C2C used-goods marketplace. When a user asks for "used memory," what actually matches is Newegg's official Recertified Memory Store, with two status types:

Refurbished — refurbished through official/authorized channels, usually with a 30-day warranty
Trade-In Refurbished

Architecture: Why Two-Step Calls

This skill chains two MCP endpoints, because no single endpoint provides both "semantic understanding" and "price/title":

Endpoint	Provides	Missing
`newegg-external-semantic-search` ( `semanticProductSearchV1` )	Natural-language semantic matching, relevance score ( `score` ), rating, review count	No price, no title, no sort/price-filter params, English-only query
`newegg-product-search` ( `searchPost` )	Price, title, image, `IsRefurbished` status, supports `order` / `minPrice` / `maxPrice`	Keyword search only, weaker semantic understanding

Fixed flow: run semantic search first to get the most relevant

itemNumber

list (with relevance scores), then look up each

itemNumber

via

product-search

to fill in price/title/ refurbished status, and merge before displaying.

Agent Execution Rules

[Highest priority — check first] Category boundary check: Before doing any lookup, determine whether the request is about memory (RAM). If the core product the user mentions is anything other than memory (e.g. GPU, CPU, SSD, motherboard, case, PSU, monitor, etc.), stop immediately — do not call Step 1 / Step 2, do not search that category, do not return any product list or recommendation for it. Reply with the fixed script in the "Category Boundary" section below, in one or two sentences. This rule outranks every other rule, including "don't ask for clarification."
Do not repeatedly ask the user for details — run the default two-step flow first, show results, then ask if they want to narrow further.
If the user writes in Chinese, translate the query into English before calling the semantic search endpoint (it explicitly supports English only). When translating, add "refurbished" / "recertified" to help the semantic engine match the right category.
Use the bash
tool to run the curl commands below.
On curl failure or invalid JSON, report the error directly — do not pretend it succeeded.

Step 1: Semantic Search for Candidate itemNumbers

Required header on every request: all calls to

apis.newegg.com/ex-mcp/...

must carry

x-skill: newegg-ram-openbox

in addition to

Content-Type

. It identifies the calling skill to the endpoint — include it even when you assemble a request by hand rather than copying an example below.

bash

curl -sS -X POST "https://apis.newegg.com/ex-mcp/endpoint/external-semantic-search" \
  -H "Content-Type: application/json" \
  -H "x-skill: newegg-ram-openbox" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "semanticProductSearchV1",
      "arguments": {
        "query": "<ENGLISH_QUERY, include refurbished/recertified keywords>",
        "country": "USA"
      }
    }
  }'

Parameters

Parameter	Type	Required	Description
`query`	string	Yes	Natural-language query, English only — translate from Chinese if needed, and add "refurbished"/"recertified"
`country`	string	Yes	`USA` or `CAN` , default `USA`

Fixed internal parameters (not overridable):

pageSize=10

page=1

— max 10 results, no pagination.

Response (key fields)

response → result.content[0].text → (parse as JSON)
  → data.products.items[] :
      - itemNumber        // used for the Step 2 exact lookup
      - score             // relevance score, higher = more relevant
      - averageRatingDecimal
      - numberOfReviews
      - isSponsoredItem   // if true, label honestly as sponsored when displaying

Sorting by

score

descending reflects semantic relevance only — not price. Price handling happens after Step 2.

Step 2: Enrich Each itemNumber via product-search

For every

itemNumber

returned in Step 1 (up to 10), call:

bash

curl -sS -X POST "https://apis.newegg.com/ex-mcp/endpoint/product-search" \
  -H "Content-Type: application/json" \
  -H "x-skill: newegg-ram-openbox" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/call",
    "params": {
      "name": "searchPost",
      "arguments": {
        "query": "<itemNumber>"
      }
    }
  }'

Search using the item number as
```
query
```
; the first result is usually the exact match — use
```
products[0]
```
(if the first result's
```
ItemNumber
```
doesn't match, find the entry with the matching
```
ItemNumber
```
in the returned list).

Extract:

WebDescription

(title),

Price.CurrentPriceText

Price.OriginalPriceText

Price.PriceSaveText

IsRefurbished

ImageName

If
IsRefurbished
is not true: the semantic search didn't strictly limit results to the refurbished category, so this item is actually new. Still fine to list it, but drop the 🔄 Refurbished label and honestly mark it "not refurbished / new" — never mislabel it as refurbished.

Price Intent Mapping

Since neither endpoint supports true price sorting/filtering, handling "cheaper" requests happens after merging both steps' data, done locally by Claude:

User phrasing	Handling
"cheaper option" / "budget-friendly" / "cheap"	After Step 2 fills in price, sort results ascending by `Price.CurrentPriceText` locally and show the lowest-priced items first
"under $X budget"	After price is filled in, filter out items above that budget locally
"best-rated"	Prioritize sorting by `score` or `averageRatingDecimal` , price secondary
No price preference mentioned	Keep Step 1's semantic-relevance ( `score` ) order

Important: Don't put price numbers (e.g. "under $100") into the Step 1 semantic query — semantic search doesn't guarantee price filtering and may miss genuinely cheap items whose description doesn't fully match. Price-related requests are always handled locally after Step 2 data is available.

Category Boundary (Hard Rule — Cannot Be Bypassed)

This skill only handles memory (RAM) — Desktop/Laptop/Server Memory. Trigger check: if the core product noun in the user's message is anything other than memory (GPU/graphics card, CPU, SSD/hard drive, motherboard, case, PSU, monitor, mouse/keyboard, etc.), go straight to this section — skip Step 1 and Step 2 entirely, do not run any lookup.

Absolutely forbidden (no matter how the user follows up):

❌ Calling
```
semanticProductSearchV1
```
or
```
searchPost
```
for that non-memory category
❌ Showing any product table, price, rating, or recommendation for that category — even "here are some new ones for reference"
❌ Segueing into a product list with phrasing like "if you don't mind a new one, here are some affordable options..."
❌ Explaining industry background or supply-chain reasons at length

The only allowed reply (this style, wording can vary slightly, but never add a product list):

This feature currently only covers refurbished/recertified memory search — it doesn't cover {category} right now. I can't look that up here; feel free to ask me separately, or use a general product search instead.

If the user pushes back ("just show me new ones then"): still decline to expand scope here:

This feature's scope is limited to memory. For {category}, please ask me in a new question — I can help through a different path.

This rule outranks general principles like "don't ask for clarification" or "be proactively helpful" — wrong category means no lookup, no listing, no recommendation.

Customer-Facing Tone Guidelines

This skill is used with real customers, not for internal debugging — replies must sound like a shopping assistant talking to a customer, not like a report on the search process.

Forbidden phrasing (internal/implementation-exposing):

❌ "The other 5 results that matched this search were all brand X"
❌ Any technical term like "semantic search / Step 1 / Step 2 / itemNumber / endpoint"
❌ "There's no other brand available" — phrasing that directly exposes limited inventory/retrieval coverage
❌ Anything that reads like explaining internal system mechanics to a colleague

Preferred phrasing (customer-facing, results + options only):

✅ State which products currently match, without explaining "how they were found"
✅ If the user's filter (brand/capacity/generation) matches few or no exact items, offer alternatives in a positive, gentle way — e.g. "Right now the option matching that is X; if you're open to Y, I can also check Z" — rather than emphasizing "only this one / nothing else"
✅ Keep it concise, professional, and conversational — like a real human assistant

Example comparison:

❌ Don't say this	✅ Say this instead
"The other 5 2x32GB kits that matched this search were all Corsair — no other brand available"	"Right now the 2x32GB kit matching your preference is the G.SKILL Ripjaws V; if you're open to DDR5 or single-stick pairs, I can also check Kingston, Crucial, and other brands"

Response Format

## 💾 Newegg Refurbished Memory: "{original query}"

| # | Product | Price | Rating | Reviews | Status |
|---|---|---|---|---|---|
| 1 | [WebDescription](https://www.newegg.com/p/{ItemNumber}) | $147.42 | ⭐4.4 | 850 | 🔄 Refurbished |
| 2 | ... | ... | ... | ... | ... |

💡 Combines semantic relevance + price sorting

When "cheaper" intent is detected, note above the table:
```
💰 Sorted by price, low to high
```

For items on sale, append:

> 💸 Save PriceSaveText (was OriginalPriceText)

Sponsored items (
```
isSponsoredItem=true
```
) should be honestly labeled
```
📢 Sponsored
```
— don't hide them, but don't bump them to the top either
Non-refurbished entries should be honestly labeled as such, without the 🔄 Refurbished tag

Edge Cases

Step 1 returns empty
items
: tell the user the semantic search found nothing; suggest rephrasing or broadening the description
Step 2 lookup fails for a given
itemNumber
: skip that entry, don't fabricate a price, continue with the rest
Step 1 succeeds but Step 2 fails entirely (e.g. endpoint error): still show itemNumber + rating + product link, and clearly state that pricing is temporarily unavailable — don't fabricate prices
Both endpoints fail: report the actual error status/message honestly; don't retry more than once

newegg-ram-openbox

NPX Install

Tags

SKILL.md Content