- Brand
- Findlane
- Category
- Commerce
- Primary Subcategory
- Ecommerce Operations Platforms
Integration details
Description
Findlane is hosted search for online stores and sites: products, docs, articles and listings. This plugin lets ChatGPT manage your Findlane search as you, in the workspaces you choose. Set up search: create an index, load records, and apply the settings Findlane suggests for your kind of data, then try searches, filters, facets and sorts to check the results. Tune it: synonyms, query rules, typo tolerance, ranking, merchandising of category pages and searches with banners, lists of products for your pages, and A/B tests of a settings change on a share of shoppers. Before a change is saved, compare_settings runs your search tests and your most searched queries with both settings and shows what moves. See how it's doing: top searches, searches that found nothing with a reason and a suggested fix for each, clicks, add-to-carts and orders from search, revenue from search, sales by filter value, and the most viewed products. Keep it safe: allowed sites for browser search, limits on what browsers can see and how fast they can search, trusted addresses, and API keys for your servers. The connection has your role in each workspace: owners can change everything, and other members can read settings, search, and see analytics. It never reaches billing, members, or workspace settings, and you can disconnect it at any time under Connected apps on a workspace's API keys page. You need a Findlane account (findlane.dev).
- Integration type
- Plugin
- Verification status
- Not applicable
- Platform
- ChatGPT
- Primary Subcategory
- Ecommerce Operations Platforms
- Secondary Subcategories
- None listed
- Brand
- Findlane
- Access
- Account required
- First tracked
- 2026-11-20
- Tool count
- 70
- Market
- US
The Primary Subcategory used for this profile’s headline score.
Other Subcategories where the Integration is listed.
Not discoverable in ChatGPT yet
ChatGPT now suggests Plugins on its own when they match a user's request, but not every Plugin is being surfaced yet.We haven’t seen Findlane suggested yet. Sign up and we’ll track it, and your Discoverability Score will appear here once ChatGPT starts surfacing it.
What discovery looks like

Get alerts for Findlane
Get updates when Findlane’s Discoverability Score or category rank changes.
Competing in ChatGPT Ecommerce Operations Platforms
View Category70 tools agents can invoke
Adds 1 to 100 records, or replaces those with the same objectID whole (send every field). Up to 256 KB a call, all or nothing. For larger loads, use create_upload_key.
upsert_records
Adds the fixes get_no_result_fixes found for these queries to the index's settings, as ordinary settings: a one-way synonym from each query to its fix.synonym, joined to the query's one-way synonym line when it has one. Queries without a fix are skipped; applied lists those added. Search each query afterwards to check it finds what it should.
apply_no_result_fixes
Sets an index's related-results floor again from its own searches that found nothing in the last 30 days: for each, the first related result it would show is judged useful or poor, and the floor becomes the lowest similarity at which at most about 1 in 20 shows a poor one (0.55 with fewer than 20 such searches). Saves it as relatedResults.minScore and returns the calibration, with examples at the floor. Related results calibrate on their own once every record is embedded, so run this only after the catalog or what shoppers search has changed a lot. Needs related results on and every record embedded, and runs at most once a day (get_related_results says when).
calibrate_related_results
Changes a key's name, scopes (the same as create_api_key's, replacing the key's), the index it's limited to (null: every index), or its searches a minute (600, 3000 or 12000; 3000 by default), a ceiling at the edge for a leaked key. Give only what changes. The secret stays the same, so servers using it don't change anything; new scopes, index and limit apply within a minute.
update_api_key
Changes any search setting: searchable fields and weights, facets, filters, sorts and the default sort, rank first, custom ranking and relevance tolerance, synonyms and one-way synonyms, query rules, typo tolerance with its word lengths and exact words, did you mean, compound words, related results, query suggestions, languages and Icelandic word forms, learning from clicks, text normalization. Only the settings given change. Lists take an array to replace them, or { add, remove } to change items and keep the rest; add also updates an item with the same field or sort ID. Returns how the change applied: instant, fields, searchable or rebuild (the last two may continue in the background; see wait_for_settings). On an index in use, run compare_settings with the same input first, and look at search tests that would fail and queries that would find fewer results before saving; after fixing a search, save it as a test with set_search_tests so later changes keep it working.
update_settings
Reads a feed's first 2,000 items without saving anything: its format (xml, tsv, csv, json or jsonl), the records those items make as they'd be stored (sample, the first five), the fields they have, and refused items with their reasons. Use it before connect_feed, and to see why items are refused. Without url, the index's connected feed. Google's and Meta's attribute names are mapped to title, url, image, images, price, compare_at_price (while a sale_price is on), currency, available, brand, product_type and categories (from product_type's levels), gtin, mpn, specs (from product_detail) and the rest under their own names; JSON files are records as the API takes them.
check_feed
Compares a settings change with the index's current settings, without saving it: runs the index's search tests, the queries given, and unless searched is false its 100 most searched queries of the last 30 days, each with both settings, for up to 20 seconds (truncated: time ran out first). Takes the same settings input as update_settings; with no settings it runs the tests with the current settings. Returns summary (how many queries' top 10 changed, found more or fewer results, now find nothing or now find something, with their searches), every test with before and after (whether it passes with the current settings and with the change; skipped when time ran out, testsNowFailing names those that pass now and wouldn't), and the 20 changes that matter most: each query's results before and after, and the results that left, joined or moved within its top 10 (places in the first 20; null is outside them). Run it before update_settings on an index in use, and don't save a change that fails tests or makes searches find fewer results without a reason: a rule that looks redundant can be what makes a search find anything, and a rule with placeholders can make searches for values no record has find nothing. Query-time settings compare: rules, synonyms, one-way synonyms, typo settings, exact words, rank first, custom ranking, relevance tolerance, languages, compound words, weights and did you mean. A change to what the index stores (new searchable, facet, filter, sort or location fields, or normalization) is refused: save it on its own first. Compares don't count as searches.
compare_settings
Connects an index to a product feed, or changes its setup, and reads it now when it's new or its address or variants changed. Each field given replaces the current one. Runs then follow the schedule, replacing the index's records with the feed's: only products that changed are written. A run that would empty the index, remove more than half its products, or refuse a fifth of its items is held, and the owners and admins are emailed; import it anyway with run_feed's held. On a first import into an index with default settings, settings for the feed's fields are applied. paused: true stops the runs; false resumes them, as after a week of failures. API writes still work between runs, and the next run replaces them.
connect_feed
Adds a subindex to an index: a small index searched with it, such as a store's categories and pages beside its products. Up to 5 per index. It's named index.name (products.categories) and works as any index in every other tool: load records, change settings, search it. Browser and API-key searches of the index ask for it with subindexes: { name: { limit } }, and count once with it. It isn't counted among the plan's indexes; its records share the index's per-index limit, and the index's keys open it.
create_subindex
A key for servers: search, write, events (sending views, clicks and orders; write keys may too), analytics (reading the index's analytics: searches, events and product counts; no other scope does), or several, optionally limited to one index. The secret is shown once. Put it in an .env file that isn't committed, never in browser code.
create_api_key
Creates an empty index. Next, load records (upsert_records, or create_upload_key for more than 100), then read get_index for suggested settings and apply them with update_settings.
create_index
A write key for one index that expires in an hour, with commands that upload a JSON file of records from your shell or a script. Use it to load more than 100 records without passing them through this conversation.
create_upload_key
Deletes one of an index's lists. Pages that ask for it get a 404 within seconds. Returns the lists left.
delete_list
Deletes a key; requests with it stop working within a minute.
delete_api_key
Deletes the index, its records, and the API keys limited to it. This can't be undone.
delete_index
Deletes an uploaded image by its key (from list_images). Refused while a rule's banner shows it.
delete_image
Deletes up to 100 records by objectID.
delete_records
Disconnects the index's product feed: its runs stop and their history goes. The index keeps its records.
disconnect_feed
Ends an index's running A/B test. keep "a" keeps today's settings; keep "b" applies the tested change to them. The results are kept as they stood. After a test ended keeping today's settings, apply: true applies its change instead.
end_ab_test
A product's recommendations, for indexes with the recommendations add-on: products that go with it (goesWith, with the decision model's probability as score), products a shopper could choose instead (similar), and products bought in the same orders (together, with count: the orders in the last year that had both), as a product page gets them. ready is false until its lists are built. Text is cut at 300 characters.
get_recommendations
One workspace: its ID and name, your role, the plan and its limits, usage this month, and its indexes.
get_workspace
Everything about one index: its settings, every field in its records with the field's type and how many records have it, suggested settings for the index's kind of data until its settings are first changed, the record count and indexing revision, a settings change in progress, browser search access (public ID, allowed origins, policy, and rate limits and rules for crawlers, scripts and data-center networks), and optimized images (allowed image hosts, the hosts found in its records, placeholder progress, and the month's new copies against the plan's).
get_index
The A/B test running on an index and the ones that ended, newest first. For each test: what it changes, group B's share, and when it started and ends or ended. outcome is "a" (kept today's settings), "b" (applied the change) or "guard" (ended early because group B's searches found nothing far more often). results compares the groups per search: searches, the share that found nothing, clicks, the share of clicks on the first 3 results, add-to-carts, orders and revenue. Each figure has group A's and B's values and B's change with a 95% interval (null until there are two days of data). decided is true once the test has run its days with 1,000 searches in each group. verdict is then "b" (B ahead), "a" (A ahead) or "none", judged on clicks per search, then add-to-carts. Before that the numbers are too early to act on.
get_ab_test
What learning from clicks does for an index (the learnedRanking setting): whether it's on, when its lists were last built (from the last 30 days), from how many clicks and credited sales, and for each search with a list, the products it ranks first within their rankFirst group, best first, with their titles, clicks, add-to-carts and orders. Searches and products in exclusions are left out; change them with update_settings learnedExclusions. holdout compares the last 14 days of shoppers searching with learning and the share held out without it each day (searches, clicks on search results and on the first 3, add-to-carts and purchases); null before any.
get_learned_ranking
Where an index's related results stand (the relatedResults setting: when a search's words find nothing, the records closest to it in meaning, in search's related). Returns whether they're on, and whether the index can have them (available, else reason: plan for a free plan, size for more than maxRecords records); embedding: the records embedded so far and those still waiting, and paused (budget once the month's budget is spent, plan, size) when it has stopped; floor, the similarity a record needs to show, null until calibrated, and none show until then; the last calibration: when it ran, from how many of the index's searches that found nothing in the last 30 days (deadEnds), the shares whose first related result was judged useful and poor, and examples at the floor (a query, how often it was searched, the first related result's title and score, and whether it was judged useful); calibrateFrom, when calibrate_related_results may run again; and spend: this month's cost in dollars against the budget, 3% of the plan's price.
get_related_results
Each banner's numbers over the last 7, 30 or 90 days, by its rule: shown (shoppers' searches that showed it, cache answers included; not crawlers or scripts) and clicks (from React's Hits, the Shopify app, or bannerClick events), for its click rate. analytics: off or unavailable when they can't be read.
get_banner_stats
Views, clicks, add-to-carts and orders the workspace's sites and servers sent, over the last 24 hours, 7, 30 or 90 days, for the workspace or one index, from browsers, servers or both: totals and the previous period's, per index, and per day (per hour for 24h, UTC). droppedFromBots counts events from crawlers and scripts, which aren't recorded; refused counts events refused for the field that was wrong (type, objectID, position, query, kind, list, from, items, quantity, price, currency, orderID, timestamp; cancel is a cancel sent from a browser). Counts are estimates from a sample, about a minute behind, kept 3 months. With recent, also the latest 20 events of the last day, with what each carried: check it after setting up events on a site, to see they arrive. Exact counts per product are get_product_analytics'. How to send events: read_docs events.
get_event_analytics
A facet field's values as facet matches see them, to go through before hiding some: the values with the most records first (a field's 2,000 with the most match), each with its records, its name and where that comes from (records: read beside the value; written: in the setting's labels; value: none found, so the value itself), whether the setting hides it, and whether it matches. Hide or show values with update_settings (facetMatches.hide, facetMatches.unhide); try queries with search and facetMatches: true.
get_facet_matches
One category page's or search's arrangement (status none when nothing is arranged), and its first products as shoppers see them now, with every rule applied: each one's place, objectID and title, and whether the arrangement placed it. Use it before set_merchandising, to choose products and see the result after.
get_merchandising
Why an index's searches found nothing, over the last 7 or 30 days, and a fix when settings can help: each query that found nothing without filters at least 3 times, more often than it found something (up to 50, the most such searches first, in searches; queries typed on the way to a longer one left out), with its cause. typo: a typo for fix.synonym (did you mean's correction); another-word: the records call it fix.synonym. Either way the fix is a one-way synonym from the query to fix.synonym, with which the query finds fix.results. not-sold: the store doesn't seem to sell it. junk: not a real search. unknown: nothing sure enough. Causes come from a decision model choosing among did you mean's correction, the index's own facet values and searches that found results, shown what each finds; a fix is offered only when very likely (confidence is how likely, 0.9 or more) and when searching the query with it added finds something. not-sold is a guess, right about 4 times in 5 when measured. cause null with pending: true means reviews are still running: call again in a minute. Reviews are kept for a week or until the records change by a tenth. appliedAt says when a fix was added. Add fixes with apply_no_result_fixes, after checking them.
get_no_result_fixes
An index's products over the last 24 hours, 7, 30 or 90 days, from every event its site and servers sent (exact counts, kept since 9 October 2026, about 2 minutes behind): for the 20 most viewed products, views, viewers (different visitor IDs, when the index takes them), clicks on search results, add-to-carts (a cart counts for each product in it) and stored orders. viewedNotBought lists products with 50 or more views and the fewest orders per view, which may need better photos, prices or descriptions; it's null when the index has no orders in the range. Empty until the site sends product views.
get_product_analytics
An index's product feed, or null: its address, schedule (hourly, 6h, or daily at hour UTC), variants setting, whether it's paused (pausedReason owner, or failing after a week of failed runs), failures in a row, when it runs next, and its last run. A run's status is imported, unchanged (the file hadn't changed), held (not imported, since it would empty the index, remove more than half its products, or a fifth of its items were refused; message says which) or failed. With runs: true, the last 20 runs, each with how many items it read, records it made, products added, changed and removed, and refused items with their reasons.
get_feed
Records by objectID (up to 20), or a few sample records to see their shape. Text longer than 300 characters is cut.
get_records
An index's revenue from search over the last 24 hours, 7, 30 or 90 days: orders and add-to-carts of products a shopper clicked in search results in the 24 hours before, by the same visitor ID (exact, from every event kept since 9 October 2026, about 2 minutes behind). Amounts are in the main currency, the one most orders use, at the prices the events carried; each order counts once. Totals beside all orders, the 30 queries with the most revenue (searches when among the 1,000 most searched; typed lists shorter queries typed on the way, folded in), and the 20 products with the most revenue from search with the query that found each most. Since 26 October 2026, a search an ad, an email or another site's link opened (the shopper didn't search) isn't counted as from search: its orders are in opened, by kind, and openedRevenue on queries and products. Orders sent as cancelled or refunded (cancel events) are left out of every number and counted in cancelled. Empty, with coverage saying why, until the index takes visitor IDs (Link each shopper's events) and its site sends search clicks and purchases with prices.
get_search_revenue
How often an index's query rules fire, over the last 24 hours, 7, 30 or 90 days: for each rule, disabled ones too, the searches whose query its phrases match, with its five most searched such queries. It's counted from the index's most searched queries against the rules as they are now, so it follows a change at once: check a rule here before changing it, to see what it catches, and after, to see that it fires on the queries meant and no others. A rule with no searches matches no real query: try other phrases, or match contains. A rule with counted: false fires on filters or contexts (or has no phrases), which searches' analytics don't keep, so it isn't counted: check it with search, sending its filters or context. queriesCounted and searchesCounted are what was looked at.
get_rule_hits
Which filter values lead to sales, for an index over the last 24 hours, 7, 30 or 90 days: for each facet field (such as brand or price), the values shoppers filtered on, with the clicks on results while each was chosen and the add-to-carts, orders and revenue of the products clicked, attributed as get_search_revenue is (a product's last click, by the same visitor within 24 hours). Searches (search) and category pages (filters without words: page, where the page's own filter counts too) apart; unfiltered is searches without filters, to compare with. A click with several values counts for each, so a field's rows can add up to more than its sales. Searches an ad, an email or a link opened are left out. From 26 October 2026 (since), when clicks started carrying their search's filters; needs the index's visitor IDs.
get_sales_by_filter
Searches over the last 24 hours, 7 days or 30 days: totals, top queries, queries that found nothing, and median time, for the workspace or one index. rising lists searches that rose most against the period before (searched at least twice as often as its traffic would lead one to expect, with the period before's searches in previous, and how many found nothing): new products, a season, something in the news; one that finds nothing is demand the store may not meet. Each query in noResultQueries (an empty one for searches with filters and no query) has what its searches filtered on (filter fields with the index), their sorts, and for browser searches from 12 October 2026, the filters with their values. traffic says who searched (visitors, verified crawlers, other crawlers, scripts, your servers, and from 30 October 2026 signed-in visitors: searches with signed keys, class signed), how many counted toward the plan or were free (cache, verified crawlers), and how many were refused and why, which answers whether a jump in searches is bots. traffic.heaviestVisitors lists the browser visitors who took the most results in a day (from 29 October 2026), each a key that's new each day and on each index, with their network, country, traffic class, searches, results returned (repeats counted) and the furthest result reached: a visitor taking about a whole index in a day may be copying it. Today's visitors on the heavy-visitor list have today (blocked or slowed for the rest of the day UTC, or unblocked) and their key, which unblock_visitor takes; you marks the visitor whose address this request comes from, and signedIn a person a signed key named, counted by them rather than their address. Each reason in refusedBy (key: a signed key refused, as expired, not made with one of the workspace's signing keys, or not for the index, site or address) has what its refused requests had in common: sites (with the index; a refused origin from 8 October 2026 whole, ready for allowed origins), countries, traffic classes, filter fields and sorts, what their clients were told, and when the latest was. Once the site sends clicks, each top query has its clicks and the middle click's place (middle, from 1), and clicks has whereClicksLand, clickedFurtherDown and searchesWithoutClicks. whereClicksLand splits clicks on search results by whether the query's results fit on one page (pageSize results; null when the indexes' differ), by its typical result count: on onePage searches shoppers look for something specific, so topThree, the share of their clicks in the top 3 (0 to 1, with previousTopThree for the period before), shows how well ranking works; morePages searches are browsing. Each group has its clicks by band; unsized clicks are on queries whose result count isn't known. clickedFurtherDown lists one-page searches whose middle click is after place 3, the most clicks after place 3 (below) first, each with the result shoppers click most after the top 3 (result, with its title and middle place; null title: no longer in the index), which may belong higher: suggest a rule that boosts it, after checking the search with explain. searchesWithoutClicks are queries with results nobody clicked, a sign the best result isn't shown first. facetMatchesPicked lists the facet matches shoppers picked, with the searches each was picked from.
get_search_analytics
An index's search tests: searches that must keep working, each a query with optional filters and what its results should hold (a record in the top N, a record kept out of the top N, or at least N results). compare_settings runs them; with no settings, it runs them with the current settings.
get_search_tests
What has come into the index, as its Updates page shows it: when data last arrived and when it last changed anything, each source (API keys by name, the dashboard, apps over MCP, the product feed, Shopify, the WooCommerce plugin) with the records it sent and how many were added, changed, unchanged and removed, writes refused with their reasons and an example objectID, replacements being sent or committed, and whether searches have caught up (indexedRevision, lagMs). Use it to check what an upload did, or why records aren't updating. Numbers start on 6 November 2026; series lists only the points when something came in.
get_writes
Limits what browser searches can return, search, filter, facet and sort by. returnFields is required: the fields browsers get, as a page shows them (a parent such as variations returns the fields inside it). Fields a page asks for that the policy hides are left out of its results; the rest still come back. Leave out searchFields, filterFields, facetFields and sortIds (the usual choice) and those follow the index's settings, so only the fields returned are limited; give any of them to limit it too (defaultSort comes with sortIds). location: true lets browsers search near a point or inside an area, which shows roughly where records are. null removes the policy, so browser search returns every field.
set_public_policy
The workspace's API keys: names, scopes, the index each is limited to, and its searches a minute. Secrets are never shown again.
list_api_keys
An index's lists: named rows a store's pages show, such as Popular, New arrivals or Halloween is coming (products, from a filter and a sort), or a menu's themes or categories (a facet field's values, with how many products each has). For each list: its settings, how many products or values it holds, and why its last build failed, if it did; with holds: true, also what each holds now, as a page gets it (text cut at 300 characters). Lists are built in the background, within about a minute of a change to the index, and served from a cache near the shopper, so they aren't searches and aren't counted. Pages read them with GET https://api.findlane.dev/api/public/indexes/<publicId>/lists/<id> (or ?ids=a,b for several). cap is how many lists the plan allows per index.
list_lists
The images uploaded for an index's banners, newest first: each one's url (https://images.findlane.dev/…, for a banner's image), key, size, type, width and height, and the rules showing it (usedBy); and what they use of the index's limits (50 images, 100 MB, 2 MB each). An image no rule shows is deleted a week later.
list_images
Every index in the workspace, with its kind of data, its icon and color in the dashboard (a null icon shows the kind's), its record count, whether browser search is on (it has allowed origins), and its subindexes (create_subindex) by full name with their record counts.
list_indexes
An index's arranged category pages and searches, as the dashboard's Merchandising page lists them: each one's target (field and value for a category page, or q for a search), its rule, order (the products placed first, in order; null when its pins have gaps, listed in pins), hidden products, banner, active times, and status: live, scheduled, ended, or off (its rule is turned off).
list_merchandising
The workspace's signing keys (ssk_…), with which its servers sign short-lived keys for logged-in visitors' browser searches: names, their first characters, the indexes each signs for (null: all), and when each was made. The keys themselves are shown once, in the dashboard, and no tool makes one: an owner or admin makes them on the API keys page. read_docs security explains signed keys.
list_signing_keys
The workspaces this connection reaches: each one's ID and name, your role, the plan, and its indexes. Start here.
list_workspaces
What an index's query suggestions offer shoppers: for q, the searches suggested as they type it (typed is where what they typed is in each, UTF-16 offsets); with q empty, the popular searches an empty search box shows. Also the list's last build: when, how many entries came from shoppers' searches (searched) and from the catalog, the catalog fields whose values it used, and an error if the last build failed. Lists are rebuilt daily for indexes searched in the last week, and within minutes of a settings change (update_settings querySuggestions); rebuild_query_suggestions rebuilds one now.
preview_query_suggestions
One page of Findlane's docs as Markdown. Pages: index, quick-start, import-from-algolia, algolia-compatible-api, records, settings, search, images, subindexes, lists, recommendations, events, analytics, security, data, set-up-search, products, help-centers, articles, listings, shopify, woocommerce, feeds, search-box, agents, browser-sdk, react, server-sdk, api, limits. "index" is the overview; "agents" covers this MCP server; "data" says what a record should be, with a recipe for each kind of data in "products", "help-centers", "articles", and "listings"; "set-up-search" sets an index up for the searches people type, with search tests and prompts to start from.
read_docs
Builds an index's query suggestion list again now, from its recent searches and its catalog, rather than at the daily rebuild, then answers as preview_query_suggestions does. What an index's query suggestions offer shoppers: for q, the searches suggested as they type it (typed is where what they typed is in each, UTF-16 offsets); with q empty, the popular searches an empty search box shows. Also the list's last build: when, how many entries came from shoppers' searches (searched) and from the catalog, the catalog fields whose values it used, and an error if the last build failed.
rebuild_query_suggestions
Findlane ChatGPT Plugin FAQ
How the directory, categories and Discoverability Score work.
Read the methodologyHow do I improve Findlane's ChatGPT Plugin discoverability?
The levers are the listing surface agents actually read: names, descriptions, keywords, tool metadata, and registry health. Which lever matters depends on where discovery breaks, which is what continuous measurement shows.
What are Findlane alternatives on ChatGPT?
As of 2026-11-20, Findlane competes with Run With AI, Shopify, 1stPage, 3DMockups, AfterShip Post-Purchase, AIOD Discounts & Free Gift, auction.fish, AutoFire and 92 more in ChatGPT Ecommerce Operations Platforms, ranked by public Discoverability Score.
Where is this profile measured?
This profile uses the geography attached to the latest public registry snapshot: US. Locale tags are intentionally omitted.