Tools
Pass a keyword and one review store to get_reviews. Read the default 10-row, 90-day window, the matched listing, and the REST mode that OpenAPI still omits.
get_reviews on Trends MCP returns the matched store listing plus newest reviews for one keyword. Valid sources are App Store, Google Play, and Amazon. Omit limit and the source prints up to 10 reviews over a 90-day window. Max limit is 20. Max days is 365. Out-of-range values clamp. They do not 400. REST mode is get_reviews. The alias reviews also works. OpenAPI 3.1 fetched 2026-11-08 names get_reviews in the path description and in a request example, but it still has no GetReviewsRequest schema and the discriminator mapping still omits the mode. One listing is matched per source. Read subject.id before treating the stars as the app you meant.
From the Trends MCP pull of 2026-11-08, a get_reviews call that sent source App Store and keyword notion with no limit, days, or country field returned window_days 90, count 10, and requests_charged 1. Rank 1 was App Store review 14752198455, title "Used to be good.", published 2026-11-05T02:25:32Z, 2 stars, helpful 0, reply null. REST with limit 1 printed count 1 and that same id.
The ceiling is 20, same as get_context, not a 400. MCP limit 21 on App Store printed count 20. REST limit 0 printed count 1. The 20-row App Store window for notion had summary.average_rating 2.55: eight 1-star rows, five 2-star, one 3-star, zero 4-star, six 5-star. replied_share 0.0. negative_share 0.65. One of those 20 rows, review 14663973234, carried helpful 1. The other 19 carried 0.
days clamps the same way. MCP days 366 with limit 1 echoed window_days 365 and still returned review 14752198455. REST days 0 echoed window_days 1 and printed count 0. REST days 1 also printed window_days 1 and count 0. Rank 1 on the 90-day pull was three days old on 2026-11-08. A 1-day window misses it. Treat 365 as the ceiling. Do not treat 366 as a documented 400.
offset is not a pager. REST offset 10 with limit 1 still printed review 14752198455. There is no second page in the payload. Lower limit if a shorter list is needed.
source all with limit 1 billed 3. App Store count 1, Google Play count 1, Amazon count 0. Amazon still returned a sources object and a matched subject, so it counted. The MCP description says only data types that answer are charged. An empty items array is still an answer. get_usage during this run printed Business 1,673 of 25,000 used at the start, then 1,700, reset 2026-12-01, cost_rule 1 request per data type in a call; get_top_trends: 1 request per 10 rows per board (limit defaults to 100). Only data types that answer are charged.
get_reviews items share kind, id, url, title, text, rating, author, published, helpful, reply, and extra. The listing sits on sources[].subject. The window math sits on sources[].summary. Those two ratings are not the same number.
On 2026-11-08, App Store for notion matched Notion: Notes, Tasks, AI, id 1232780281, subject.rating 4.8 from 90,639 ratings, developer Notion Labs, Incorporated, genre Productivity, version 1.7.352. Storefront URL:
https://apps.apple.com/us/app/notion-notes-tasks-ai/id1232780281
``` The 10-row window had `summary.average_rating` 2.4, with five 1-star rows, two 2-star, and three 5-star. `replied_share` 0.0. `negative_share` 0.7. Newest-first store rows skew toward people who just opened the app to complain. They are not the all-time histogram.
Google Play for notion matched `Notion: AI Workspace`, package `notion.id`, `subject.rating` 4.6 from 405,380 ratings. `limit` 1 printed `summary.average_rating` 5.0 from that single 5-star row, review `b6d3a0e3-23e0-4478-b468-5d8e7216e44a`, published `2026-11-07T02:11:02Z`, `title` null, extra `app_version` `0.6.4188`. Play rows in this pull left `title` empty. App Store rows filled it.
ChatGPT on Google Play matched `com.openai.chatgpt`, `subject.rating` 4.8 from 67,608,146 ratings. Five newest rows were all 5 stars, `replied_share` 0.0, published between `2026-11-07T06:00:35Z` and `2026-11-07T06:01:47Z`. Uber on Google Play matched `com.ubercab`, `subject.rating` 4.7 from 20,365,628 ratings. Ten newest rows had `summary.average_rating` 4.6, nine 5-star and one 1-star, `replied_share` 0.0. `reply` was null on every Notion App Store row, every ChatGPT Play row, and every Uber Play row in these pulls. The field exists. These windows did not fill it.
`insights` was null on App Store and Google Play. Amazon fills `insights` instead of `summary`. Parse per source.
## Check the matched listing before reading the stars
Each source matches the keyword to one listing and returns that listing as `subject`. The catalog says App Store accepts an app name or numeric App Store id, Google Play accepts an app name or Android package, and Amazon accepts a product name or ASIN. The live pulls matched that wording.
Keyword `1232780281` on App Store with `limit` 1 returned the same Notion app id `1232780281` and the same rank-1 review `14752198455`. `alternatives` was `[]`. Keyword `notion` on App Store filled three alternatives: Notion Calendar id `1607562761` (7,797 ratings), To-Do List & Tasks for Notion id `1594021892` (761), Widget for Notion id `6751851452` (39). An exact id pins the listing. A name search also returns near-misses.
Keyword `notion.id` on Google Play pinned package `notion.id` and also left `alternatives` empty. Keyword `notion` on Google Play filled Notion Calendar (`com.cron.calendar`), Notion: Music Notation & Tab (`com.presonus.notionmobile`), and Microsoft OneNote (`com.microsoft.office.onenote`). The music app is a different product that shares the word. Read `subject.id` before averaging stars across stores.
Amazon on the word `notion` matched Kindle ASIN `B0DJZ45JCQ`, a software book titled as a Notion operating system guide, `subject.rating` 4 from 2 ratings. `count` 0 items. The alternative was another 2026 Notion guidebook, ASIN `B0H4HWSSJB`, 1 rating. That is not the productivity app.
Country is a two-letter storefront code. Omit it and App Store used the U.S. listing, rating 4.8 from 90,639. This body printed the same url, rating, and rank-1 review:
```json
{"mode": "get_reviews", "source": "App Store", "keyword": "notion", "country": "US", "limit": 1}
country GB kept id 1232780281 but switched the url to apps.apple.com/gb/, dropped the rating to 4.7 from 15,105, and printed count 0 reviews. country USA did not alias. The payload returned subject.id null, count 0, and this note:
country must be a two-letter code such as US, GB or DE, not 'USA'.
get_top_trends aliases the three-letter form to the two-letter storefront. get_reviews does not.
Lowercase folds. REST source app store echoed App Store. REST source google play echoed Google Play. A comma list that includes YouTube fails the whole call with invalid_source. It does not drop the bad name and keep the stores.
Amazon is the third review source. It is not the Amazon search index used on get_growth. The 2026-11-08 ASIN pull for B0CJZMP7L1 matched STANLEY Quencher H2.0 Flow State Tumbler, 30 oz, Lilac, brand STANLEY, subject.rating 4.7 from 207,924 ratings, url https://www.amazon.com/dp/B0CJZMP7L1. summary was null. insights.star_split was 5: 87, 4: 7, 3: 2, 2: 1, 1: 3. Topic Quality had 12,235 mentions, 10,481 positive, 1,754 negative. Appearance had 10,891 mentions, 10,684 positive. Leakage had 2,935 mentions, 791 positive, 2,144 negative.
The one returned item had kind excerpt, id B0CJZMP7L1:Quality:0, rating null, published null, helpful null, author.name null, extra topic Quality. App Store and Google Play rows in the same morning's pulls had kind review and a timestamp. Amazon did not.
The source note was Amazon shows excerpts and topic counts without sign-in, not full dated reviews; days does not filter them. Keyword notion on Amazon printed that same note, insights.summary null, topics [], and count 0 items, with star_split 5: 49, 3: 51. window_days 90 was still in the envelope. Do not treat that window as a filter on Amazon excerpts. Do not treat count 0 as a failed source. The listing still matched.
POST https://api.trendsmcp.ai/api with this body returned the default App Store 10 on 2026-11-08:
{"mode": "get_reviews", "source": "App Store", "keyword": "notion"}
mode reviews with limit 1 returned the same rank-1 review 14752198455. A garbage mode getreviews returned 400 invalid_mode and this list: get_context, get_growth, get_reviews, get_time_series, get_top_trends. The aliases object on that error named reviews for get_reviews. It named aliases for get_time_series, get_growth, and get_top_trends. It omitted get_context, even though mode context with YouTube still returned a video the same morning.
Missing REST fields used missing_parameter, which matches docs.md. Omit keyword and the message is keyword is required: a brand, app, product or place name. Omit source and the message is source is required: one or more of App Store, Google Play, Amazon (comma-separated). Failed 400s are free per pricing.md. A successful multi-source call is the one that bills.
OpenAPI 3.1 fetched the same morning describes five tools in the path text, including get_reviews at 1 request per source. The request examples include a Get Reviews body with keyword notion and source App Store. components.schemas still has no GetReviewsRequest and no reviews response type. The mode discriminator mapping lists get_context, get_growth, get_time_series, and get_top_trends only. The 200 example set omits a reviews payload. llms.txt, docs.md, pricing.md, and data-catalog.md last updated 2026-11-08 all name get_reviews. HTML /docs documents the parameters. Call the endpoint. Do not treat the schema discriminator as the tool list.
REST get_usage is not a mode. mode get_usage returned invalid_mode. Usage stays on the MCP get_usage tool.
FAQ