Search Zepto for products matching a query at a given Indian delivery location, returning in-stock items with name, brand, pack size, price, MRP, discount, rating, review count, stock state, ETA, product URL, and image URL. Read-only.
Works with
AI-first code editor with Composer
Before installing skills in Cursor, ensure your development environment meets these requirements:
node --versionsearch-productsExecute the skills CLI command in your project's root directory to begin installation:
Fetches search-products from zepto.com/search-products-wtqzgb and configures it for Cursor.
The CLI shows a list of agents. Use arrow keys and space to select Cursor:
Confirm successful installation by checking the skill directory location:
Restart Cursor to activate search-products. Access via /search-products in your agent's command palette.
We perform automated surface-level scans (Gen AI Scanner, Socket, Snyk) during installation. These checks detect common vulnerabilities but do not guarantee complete security. Always review skill source code and verify the publisher's reputation before production use.
Skills execute code in your environment. Always review source, verify the publisher, and test in isolation before production.
Submit your Claude Code skill and start earning
Automate repetitive workflows and reduce manual effort
Example
Generate reports, summarize documents, draft communications
Save 3-5 hours per week on routine tasks
Learn new skills, understand complex topics, get expert guidance
Example
Explain concepts, provide examples, suggest learning resources
Accelerate learning and skill development by 2x
Enhance output quality through reviews, suggestions, and refinements
Example
Review drafts, suggest improvements, catch errors
Improve work quality by 30-40% with less effort
0
total installs
0
this week
0
upvotes
Run in your terminal
0
installs
0
this week
—
stars
| name | search-products |
| title | Zepto Product Search |
| description | >- Search Zepto for products matching a query at a given Indian delivery location, returning in-stock items with name, brand, pack size, price, MRP, discount, rating, review count, stock state, ETA, product URL, and image URL. Read-only. |
| website | zepto.com |
| category | quick-commerce |
| tags | - grocery - india - quick-commerce - search - read-only |
| source | 'browserbase: agent-runtime 2026-05-19' |
| updated | '2026-05-19' |
| recommended_method | browser |
| alternative_methods | [] |
| verified | true |
| proxies | true |
Given a free-text product query and an Indian delivery location (street/area/landmark name or pincode), return the best-matching in-stock products listed on Zepto for that location's delivery store — with product name, brand, pack size/variant, current price, MRP, discount, rating, review count, in-stock status, per-card delivery ETA, canonical product URL, and image URL. Read-only — never clicks ADD, never proceeds toward cart/checkout.
The Zepto web site at www.zepto.com is a Next.js SPA. The recommended path is browser-driven via a Browserbase remote session, navigating directly to https://www.zepto.com/search?query=<URL-encoded query> and parsing product cards from browse snapshot. A backing JSON API exists (POST https://bff-gateway.zepto.com/user-search-service/api/v3/search) but is gated by per-session secrets — see Site-Specific Gotchas; do not try to call it cold.
Create the session. Use a Browserbase session with stealth + proxies, region ap-southeast-1 (Singapore — closest geographically; lowest latency to Zepto's India CDN and least likely to hit IP-mismatch heuristics). Verified + proxies is required: Zepto serves a CloudFront-fronted Next.js app that fingerprints aggressively, and the address-autocomplete + map widgets fail without a residential India-routable IP.
sid=$(browse cloud sessions create --keep-alive --proxies --verified \
--region ap-southeast-1 \
| node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")
export BROWSE_SESSION="$sid"
Set the delivery location BEFORE searching. Zepto's IP-based default is unreliable (it picks an arbitrary metro and may resolve to a non-serviceable area), and price/availability/ETA all vary by storeId. The site uses Google-Places-style address autocomplete, not pincode entry — you type a free-text address and pick a suggestion.
browse open "https://www.zepto.com/" --remote
browse wait load --remote && browse wait timeout 2000 --remote
browse snapshot --remote # find ref for `button: Select Location`
browse click "[<select-location-ref>]" --remote
browse wait timeout 1500 --remote
browse snapshot --remote # find ref for `textbox: Search a new address`
browse click "[<textbox-ref>]" --remote
browse type "Koramangala Bangalore" --remote
browse wait timeout 2500 --remote
browse snapshot --remote # 5 suggestions appear under the textbox
# Click the first matching suggestion's container div ref. Suggestions look like:
# [N] div
# [N+1] StaticText: Koramangala
# [N+2] StaticText: Koramangala, Bangalore, Karnataka, India
browse click "[<first-suggestion-ref>]" --remote
browse wait timeout 3000 --remote
After the suggestion is clicked, the modal closes, the page reloads, and the header switches from button: Select Location to button: <Area Short Name> plus a heading: <N> minutes ETA badge. This is the success signal — do not search until you see that header heading.
Search via URL. The query is a URL param; no need to click into the search box.
QUERY=$(node -e "process.stdout.write(encodeURIComponent('basmati rice'))")
browse open "https://www.zepto.com/search?query=$QUERY" --remote
browse wait load --remote && browse wait timeout 2500 --remote
browse snapshot --remote > /tmp/zsnap.json
The location set in step 2 persists across the navigation — the new page header still shows the same area button and the same global ETA.
Branch on what the page shows. Inspect the snapshot tree for one of five top-level patterns:
Pattern in tree text | Outcome |
|---|---|
heading: Showing results for "<q>" followed by product link: rows | success — parse results (step 5) |
heading: Could not find any products for "<q>" | no_results — return empty array |
heading: Sit Tight! We're Coming Soon! + Our team is working tirelessly… | not_serviceable — Zepto does not deliver to the selected location |
heading: Try Again in <N> Mins (no products) | store_offline — local dark store currently closed; retry later |
heading: Try Again in <N> Mins + product rows | store_offline_but_browsable — results still rendered for browsing but ordering is paused; rating/price valid, ETA NOT shown per-card |
Parse each product card. Every result is an [X-Y] link: <inline summary> whose href (in urlMap) points to https://www.zepto.com/pn/<slug>/pvid/<uuid>. The link's accessible name is a single concatenated string in this fixed order — extract via the nested StaticText children, NOT by regexing the link name (the accessible name omits some fields and concatenates without separators):
image: <Product Name> — accessible alt = product name. The actual image URL is in urlMap[<image-ref>] and looks like https://cdn.zeptonow.com/production/ik-seo/tr:w-403,ar-...,pr-true,f-auto,q-40,dpr-2/cms/product_variant/<uuid>/<slug>.jpeg.button: ADD — present iff the product is currently in stock. An out-of-stock card replaces this with a Notify Me button or omits the action entirely. Always check for the literal button: ADD text on the card to decide inStock.image: P3 - Ad.png (or P1 - …, P2 - …) — sponsored/Premium ad slot. Treat the card normally but set sponsored: true.span → StaticText: ₹ + StaticText: <integer> — discounted (current) selling price in ₹ (whole rupees, no decimals on the site).span → StaticText: ₹ + StaticText: <integer> — MRP. Present only when there is a discount. If absent, treat mrp = sellingPrice and discount = 0.StaticText: ₹<N> + StaticText: OFF — flat-amount discount badge. The percent can be derived as (N / mrp) * 100.StaticText: <Product Name> — canonical product name (use this, not the image: alt — image: is sometimes truncated).StaticText: <pack-size> — e.g. 1 pc (20.2 g), 1 pack (500 ml), 1 L, 1 pack (5 kg). Free-form string; preserve verbatim.StaticText: <variant> — sub-variant label like Milk Chocolate, Long Grain, Silk Oreo, Fruit & Nut, Premium. Optional; may be absent on basics.span → image (star icon) + StaticText: <rating> (e.g. 4.6, 5) followed by span → StaticText: ( + StaticText: <count> + StaticText: ). Review count uses K/M suffix (e.g. 274.5k, 1.9k, 855, 125.4k). Parse with parseFloat(num) * (suffix === 'k' ? 1000 : suffix === 'm' ? 1e6 : 1). Many cards have no rating block at all — treat as null.StaticText: <N> mins — per-card delivery ETA, e.g. 9 mins. Present only when the selected store is online (i.e. the global header reads <N> minutes, not Try Again in <N> Mins).StaticText: Bestseller / StaticText: New — quality/recency tags.Brand is not a separate field on the card. Zepto inlines the brand into the product name (e.g. Cadbury Dairy Milk Chocolate Bar Cricket Pack, Nandini Goodlife Toned UHT Milk, Nestle Kitkat 4 Fingers …). Extract by matching the first word(s) of the product name against a known-brand list, OR — preferred for accuracy — open the product detail page (PDP) at the card's /pn/<slug>/pvid/<uuid> URL; the PDP returns the brand as a structured field. Only do PDP fetches for the top-N results the caller actually cares about; doing it for every card on a 263-result query is wasteful.
Pagination. Results render as a vertical-feed grid of PRODUCT_GRID widgets (3 cards per row, ~30 cards per page). To fetch more, scroll the page (browse mouse scroll 600 400 0 1500, wait 1500ms, re-snapshot, dedupe by pvid UUID). The site lazily loads more grids — there's no "Next page" button. Sponsored ad cards are interleaved between organic grids; tag them with sponsored: true (presence of P3 - Ad.png / Sponsored label) and exclude or rank lower depending on caller intent.
Release the session.
browse cloud sessions update "$sid" --status REQUEST_RELEASE
Zepto's web app calls POST https://bff-gateway.zepto.com/user-search-service/api/v3/search with body {"query":"<q>","pageNumber":0,"mode":"SHOW_ALL_RESULTS","userSessionId":"<uuid>"}. The response is a clean JSON layout with productResponse blocks containing product.brand, productVariant.formattedPacksize, productVariant.ratingSummary.{averageRating,totalRatings}, discountedSellingPrice (paise — divide by 100 to get ₹), mrp (paise), outOfStock, etc. — far richer than the rendered card. However, the request requires ~30 headers including x-csrf-secret, x-xsrf-token, request-signature (SHA hash of body + secret), storeId/store_ids (derived from the selected delivery location), deviceId, sessionId, and a 1.5KB compatible_components feature-flag string. The signature is rotated per request and validated server-side. We attempted to replay a captured request out-of-band; verified blocked. The only viable use of the API is in-page via browse eval after the same session has set its location, which is strictly more expensive than parsing the rendered DOM. Don't waste time on cold API calls.
<N> minutes before issuing search queries.Bandra West Mumbai, Koramangala Bangalore, Sector 18 Noida) and click the first suggestion. The textbox accepts free-form English; Hindi script also works but English is the safer canonicalization.tree text for the literal string Your Location). When verifying the modal opened, look for textbox: Search a new address, not for a top-of-tree dialog. The xpathMap puts the modal inside body/div[2]/div[2]/... while the rest of the page lives in body/div[2]/div[1]/..../search?query= URL. Always branch on the top-level heading: text before parsing cards. Showing results for "<q>", Could not find any products for "<q>", Sit Tight! We're Coming Soon!, Try Again in <N> Mins (alone), and Try Again in <N> Mins + products — each requires different handling. Defaulting to "parse all link: rows under banner" without checking the heading yields silent garbage on the not-serviceable and offline-store branches.Try Again in <N> Mins is per-store, not global. A given location's dark store can be offline temporarily (curfew hours, restocking, weather). Other locations the same minute may show 9 minutes fine. If Try Again in N shows with no product rows, the only fix is changing location or waiting; retrying immediately will not help.Cadbury, Nestle, Amul, Nandini, Daawat, Hocco, Tata) but breaks for multi-word brands (Mother Dairy, Pure Temptation, Bb Royal, B Natural). For accurate brand extraction, hit the PDP at /pn/<slug>/pvid/<uuid> — the PDP page title format is Buy <Product Name> Online - Price @ ₹<N> | Instant Delivery | Zepto and the breadcrumb under the buy box gives Home > <L2 Category> > <L3 Category>. The cleanest brand source is the API's productResponse.product.brand field; if you can extract it from an in-page XHR you've already paid for, do that instead of a second PDP roundtrip.discountedSellingPrice: 1800 = ₹18.00). Don't confuse the two: if you scrape the card, you have ₹; if you scrape an XHR response, divide by 100.null, not 0.P3 - Ad.png / P1 - … / P2 - … marks sponsored cards. They are real products but are paid placements. Set sponsored: true and consider de-ranking them when computing "best match".274.5k = 274,500. 1.9k = 1,900. 42.3k = 42,300. Parse before storing.urlMap are ImageKit-transformed CDN paths. They include resize/quality params (tr:w-403,ar-4000-4000,pr-true,f-auto,q-40,dpr-2). If you want a higher-res version, strip the tr:…/ segment to get the original path under cms/product_variant/<uuid>/<slug>.jpeg at cdn.zeptonow.com/production/.bff-gateway.zepto.com/user-search-service/api/v3/search is signed and gated. Headers include a rotating x-csrf-secret, a body-signed request-signature, store IDs derived from the selected delivery location, and a 1.5KB compatible_components flag set. Cold curl/fetch returns 401/403 (or worse — a 200 with empty/sanitised payload). Don't bother trying to spoof; either parse the rendered DOM, or use browse eval from within an already-authenticated session.bff-gateway.zepto.com and api.zeptonow.com are NOT the same surface. The latter returns 500 on every cold path we probed; only bff-gateway carries the v3 search API, and only behind in-page auth.*.zeptonow.com, *.zeptonow.dev, *.zepto.com, cdn.zeptonow.com. Don't try to inject third-party JS for scraping; browse eval (page-context) works because it runs as the page's own origin, but anything else will get CSP-blocked.--region ap-southeast-1 (Singapore) is closest to Zepto's India CloudFront edges and the address-suggestion latency is bearable (~2.5s for the 5-item dropdown). EU/US regions add 600-900ms of latency on every modal interaction and sometimes time out the autocomplete typeahead./search?query= snapshot returns only the first ~30 cards (10 PRODUCT_GRID widgets of 3 each). To get the full 263-result set for a popular query like "chocolate", you must scroll-paginate; there is no ?page=N URL param that works.browse open calls within the same session. Setting location once means subsequent /search?query=… navigations keep that location. This is the only reason the URL-search step is cheap; if the session restarts, redo step 2.ADD, never proceed to /cart or /checkout. Cart-modification XHRs are gated by the same signed-header machinery as search and don't easily round-trip back.Five distinct outcome shapes; the caller should branch on status:
// success — normal results
{
"status": "success",
"query": "basmati rice",
"location": {
"label": "Koramangala",
"fullAddress": "Koramangala, Bangalore, Karnataka",
"etaMinutes": 9
},
"totalShown": 30,
"totalAvailable": 263,
"products": [
{
"name": "Daawat Pulav Basmati Rice | Long Grain",
"brand": "Daawat",
"packSize": "1 pack (500 g)",
"variant": "Long Grain",
"price": 75,
"mrp": 90,
"discount": { "amount": 15, "percent": 17 },
"rating": 4.6,
"reviewCount": 16800,
"inStock": true,
"etaMinutes": 9,
"sponsored": false,
"tags": [],
"url": "https://www.zepto.com/pn/daawat-pulav-basmati-rice-long-grain/pvid/96b12acf-bdcd-47bb-a9e9-a04364fd30e1",
"imageUrl": "https://cdn.zeptonow.com/production/ik-seo/tr:w-403,ar-3000-3000,pr-true,f-auto,q-40,dpr-2/cms/product_variant/19b0b091-07e3-40b6-b60e-0719e39caf64/Daawat-Pulav-Basmati-Rice-Long-Grain.jpeg"
}
]
}
// no_results — query matched zero SKUs in this location
{
"status": "no_results",
"query": "xyzzyplugh",
"location": { "label": "Bandra C", "fullAddress": "Bandra C, Mumbai, Maharashtra", "etaMinutes": 9 },
"products": []
}
// not_serviceable — Zepto doesn't deliver to the chosen address
{
"status": "not_serviceable",
"query": "milk",
"location": { "label": "Other", "fullAddress": "Andaman and Nicobar Islands", "etaMinutes": null },
"message": "Sit Tight! We're Coming Soon!"
}
// store_offline — local dark store temporarily closed, no products shown
{
"status": "store_offline",
"query": "coffee",
"location": { "label": "Koramangala", "fullAddress": "Koramangala, Bangalore, Karnataka", "etaMinutes": null },
"retryInMinutes": 15,
"products": []
}
// store_offline_but_browsable — store paused but the catalog still renders
// for browsing. Cards have valid price/MRP/rating but NO per-card ETA;
// inStock should be reported as `null` (intent: orderable when reopened),
// not `true` — placing an order would be blocked at checkout time.
{
"status": "store_offline_but_browsable",
"query": "coffee",
"location": { "label": "Koramangala", "fullAddress": "Koramangala, Bangalore, Karnataka", "etaMinutes": null },
"retryInMinutes": 15,
"products": [
{
"name": "Nescafe Classic - Instant Coffee Powder - Pure Coffee",
"brand": "Nescafe",
"packSize": "1 pc (24 g)",
"variant": null,
"price": 116,
"mrp": 124,
"discount": { "amount": 8, "percent": 6 },
"rating": 4.7,
"reviewCount": 12800,
"inStock": null,
"etaMinutes": null,
"sponsored": false,
"tags": [],
"url": "https://www.zepto.com/pn/nescafe-classic-instant-coffee-powder/pvid/<uuid>",
"imageUrl": "https://cdn.zeptonow.com/production/ik-seo/tr:w-403,ar-1200-1200,pr-true,f-auto,q-40,dpr-2/cms/product_variant/<uuid>/Nescafe-Classic.jpeg"
}
]
}
Prerequisites
Time Estimate
15-45 minutes depending on use case complexity
Steps
Common Pitfalls
✓ Do
✗ Don't
💡 Pro Tips
✓ Use when
Use when skill capabilities match your task, clear ROI on time saved, and you can validate outputs. Best for repetitive tasks, learning, and quality improvement.
✗ Avoid when
Avoid when task requires deep expertise you can't validate, involves sensitive decisions, or when learning process is more valuable than speed of completion.
aliexpress.com/search-product-p0h8a7
kostja94/marketing-skills
aaaaqwq/claude-code-skills
agentbay-ai/agentbay-skills
shopmeskills/mcp
glebis/claude-skills
search-products is among the better-maintained entries we tried; worth keeping pinned for repeat workflows.
Useful defaults in search-products — fewer surprises than typical one-off scripts, and it plays nicely with `npx skills` flows.
Solid pick for teams standardizing on skills: search-products is focused, and the summary matches what you get after install.
Registry listing for search-products matched our evaluation — installs cleanly and behaves as described in the markdown.
search-products has been reliable in day-to-day use. Documentation quality is above average for community skills.
search-products reduced setup friction for our internal harness; good balance of opinion and flexibility.
search-products fits our agent workflows well — practical, well scoped, and easy to wire into existing repos.
Keeps context tight: search-products is the kind of skill you can hand to a new teammate without a long onboarding doc.
We added search-products from the explainx registry; install was straightforward and the SKILL.md answered most questions upfront.
I recommend search-products for anyone iterating fast on agent tooling; clear intent and a small, reviewable surface area.
showing 1-10 of 42