Ebay Hardware Hunter

eBay Hardware Hunter

An agent skill that shops eBay for hardware on a budget, scores the listings, and records them - all through the Zapier SDK. (Build walkthrough: https://mrc.fm/hardwarehunter.) The native eBay Zapier app is seller-focused (orders, fulfilment), so this skill reaches the buyer-side eBay Browse API through Zapier's authenticated curl, which injects your connected eBay account's OAuth token. Everything stays inside the Zapier SDK.

search (eBay Browse API via Zapier curl)  ->  score  ->  log (Zapier Table)  ->  report best deal + backups

Prerequisites (one-time setup)

1. Node.js 20+ and the Zapier SDK packages:

   npm install @zapier/zapier-sdk@^0.70.2 @zapier/zapier-sdk-cli@^0.55.3
   

   Every command below is run as npx zapier-sdk <command>.

2. Log the SDK into Zapier (browser auth, token stored locally for the OS user that runs it):

   npx zapier-sdk login
   

   Verify with npx zapier-sdk get-profile.

   Headless box? The login prints an authorize URL and listens on localhost:<port> for the callback. Open the URL, approve, then curl the resulting http://localhost:<port>/oauth?code=...&state=... URL from the same machine the CLI is running on to complete the handshake.

3. Connect your eBay account to Zapier at https://zapier.com/app/connections -> Add connection -> eBay -> sign in and authorize. This grant carries the base api_scope the Browse API needs. Confirm it exists:

   npx zapier-sdk find-first-connection ebay --json
   

Configuration

Read these knobs from the user's request; fall back to the defaults.

Procedure

Step 1 - Resolve the eBay connection id

npx zapier-sdk find-first-connection ebay --json

Take .data.id. If data is null, stop and tell the user to connect eBay (Prerequisite 3).

Step 2 - Search eBay (Browse API through Zapier)

Build the Browse API URL and call it with curl --connection <id>, passing the marketplace header. URL-encode the query and the filter value.

npx zapier-sdk curl \
  "https://api.ebay.com/buy/browse/v1/item_summary/search?q=mac%20mini&limit=15&sort=price&category_ids=111418&filter=price:%5B50..400%5D,priceCurrency:GBP,buyingOptions:%7BFIXED_PRICE%7D,conditions:%7BUSED%7CNEW%7D" \
  --connection <CONNECTION_ID> \
  -H "X-EBAY-C-MARKETPLACE-ID: EBAY_GB"

Notes for building the URL from config:

• q= URL-encoded QUERY.
• category_ids= CATEGORY_ID (omit the param entirely if CATEGORY_ID is blank).
• limit= MAX_RESULTS (max 200 per page).
• sort=price returns cheapest first.
• filter=price:[MIN_PRICE..BUDGET],priceCurrency:CURRENCY,buyingOptions:{BUYING_OPTIONS},conditions:{CONDITIONS}
  • encode [ as %5B, ] as %5D, { as %7B, } as %7D, | as %7C.
• Header X-EBAY-C-MARKETPLACE-ID: MARKETPLACE.

Relevance guard - do not skip this. A bare keyword search for popular hardware is full of accessories and parts. q=mac mini alone surfaces docks, hubs, stands, and even "BOX only" listings that are cheap and therefore score as "great deals". Two defences, both applied: (1) CATEGORY_ID restricts the server-side search to actual machines; (2) the EXCLUDE keyword check in Step 3 drops anything whose title still betrays a non-machine.

The response is JSON with an itemSummaries array. For each item read: title, price.value, shippingOptions[0].shippingCost.value (absent -> treat as collection/unknown, default 0), condition, buyingOptions, seller.username, seller.feedbackPercentage, seller.feedbackScore, marketingPrice.discountPercentage (optional), itemId, and itemWebUrl (the direct buy link).

Step 3 - Score each listing (0-100)

First drop irrelevant listings: skip any item whose lowercased title contains an EXCLUDE keyword (box only, dock, hub, stand, case, charger, cable, etc.). This is what stops an empty "Mac Mini BOX only" from topping the ranking.

Then compute total = price + shipping and skip anything with total > BUDGET. Sum the components below (clamp the final score to 0-100). State the rubric to the user so the ranking is transparent.

• Value vs budget - up to 40. 40 * (BUDGET - total) / BUDGET. Cheaper relative to budget scores higher; at-budget scores ~0.
• Seller trust - up to 25. 25 * (feedbackPercentage / 100), then subtract up to 10 if feedbackScore < 50 (thin track record). Floor at 0.
• Condition - up to 15. NEW/Brand New = 15; Certified/Excellent - Refurbished = 12; Used/Very Good/Good = 8; For parts or not working = 0.
• Buying option - up to 10. FIXED_PRICE (buy-it-now, instant) = 10; FIXED_PRICE with BEST_OFFER = 10 (+ note offer is possible); AUCTION = 5 (price not final).
• Shipping - up to 10. Free = 10; otherwise 10 * (1 - shipping / 20) floored at 0 (i.e. >= 20 in shipping scores 0).
• Discount bonus - up to 5. If marketingPrice.discountPercentage present, add min(5, discountPercentage / 10).

Sort listings by score descending. The top one is the recommended deal.

Step 4 - Ensure the log table exists

npx zapier-sdk list-tables --json

Find a table whose name equals TABLE_NAME. If none exists, create it and its fields (one-time):

npx zapier-sdk create-table "Mac mini deals" --description "eBay hardware hunter log"
# take .data.id from the output, then:
npx zapier-sdk create-table-fields <TABLE_ID> '[
  {"name":"title","type":"text"},
  {"name":"score","type":"number"},
  {"name":"price_gbp","type":"text"},
  {"name":"shipping_gbp","type":"text"},
  {"name":"total_gbp","type":"text"},
  {"name":"condition","type":"text"},
  {"name":"buying_options","type":"text"},
  {"name":"seller","type":"text"},
  {"name":"seller_feedback_pct","type":"text"},
  {"name":"seller_feedback_score","type":"number"},
  {"name":"url","type":"text"},
  {"name":"item_id","type":"text"},
  {"name":"found_at","type":"text"}
]'

Money and percentage values are stored as text on purpose: Zapier Table number fields round to integers on write, which would turn 319.99 into 319. Keep score and seller_feedback_score as number.

Step 5 - Log the scored listings

Each record must be wrapped in {"data": {...}}. Write all scored listings (up to 100 per call):

npx zapier-sdk create-table-records <TABLE_ID> '[
  {"data":{
    "title":"Apple Mac mini (2020) M1 256GB 8GB",
    "score":82,
    "price_gbp":"319.99",
    "shipping_gbp":"0.00",
    "total_gbp":"319.99",
    "condition":"Excellent - Refurbished",
    "buying_options":"FIXED_PRICE",
    "seller":"clove-technology",
    "seller_feedback_pct":"99.1",
    "seller_feedback_score":15282,
    "url":"https://www.ebay.co.uk/itm/267581736217",
    "item_id":"v1|267581736217|0",
    "found_at":"2026-06-15T16:45:00Z"
  }}
]'

Use the item's itemWebUrl for url. Set found_at to the current UTC time.

Step 6 - Report to the user

Reply in chat with:

• Best deal: title, total price, condition, seller (feedback %), buy link, and one line on why it won (the scoring breakdown).
• 2-3 backups: title, total, link.
• A line confirming how many listings were logged to the TABLE_NAME table.

This chat report is the "notification" (this skill is notification-channel agnostic; see Extending below to fan out to Slack, etc.).

Failure modes & fallbacks

• find-first-connection ebay returns null -> eBay isn't connected. Send the user to https://zapier.com/app/connections. Do not proceed.
• Browse API returns errorId 1100/"Access denied" or HTTP 403 -> the connected eBay token lacks Browse scope. Reconnect eBay in Zapier; if it still fails, the account may need a fresh OAuth grant. (In testing, a standard eBay connection carries the base api_scope the Browse search needs.)
• "total": 0 / empty itemSummaries -> loosen the filter: widen BUDGET, add AUCTION to BUYING_OPTIONS, or broaden CONDITIONS. Report the empty result honestly rather than inventing listings.
• Auth error on any zapier-sdk call -> run npx zapier-sdk login as the OS user that will run the skill (the token is per-user).

Known limitations

• The Browse API item summary does not expose watcher counts or bid counts, so those are not scored or logged.
• This skill finds and records deals and hands you a purchase link - it does not place bids or buy. eBay's API and the Zapier eBay app do not expose buyer bidding/checkout, so completing a purchase stays a human click. (This matches the "prepare checkout / deep link" approach.)

Extending

• Other hardware: change QUERY (e.g. rtx 4090, ddr5 64gb, thinkpad x1). Tune BUDGET/CONDITIONS to match.
• Other notification channels: any Zapier-connected app can be the alert. After Step 6, e.g. Slack:

  npx zapier-sdk run-action slack write create_message \
    --inputs '{"channel":"#deals","text":"Best Mac mini lead: ..."}'
  

  (Requires a Slack connection. Discover inputs with npx zapier-sdk get-action-input-fields-schema slack write create_message.)
• Scheduling: run this skill on a cron to hunt daily and keep the table as a running price history.