Pass one source and one keyword to get_time_series, then count the points that come back. Weekly Google series, monthly Amazon series, and app-download cumulative fields do not share a clock.
Live data as of 2026-09-03
get_time_series on Trends MCP returns the stored 0-100 series for one keyword on exactly one source. REST mode is get_time_series. Aliases are get_trends, trends, time_series, and lookup. On the 2026-09-03 pull, nike on google search printed 261 weekly points from 2021-09-04 (index 71, volume 29,221,488) through 2026-08-29 (index 64, volume 26,340,496). Amazon air fryer printed 49 month-end points ending 2026-07-31. Steam, Wikipedia, python, npm, and app downloads labeled the source as datatype. google search and google shopping used source. google shopping omitted volume and stamped dates as UTC datetimes. all and comma-separated sources returned invalid_source. App downloads added volume_cumulative; com.whatsapp printed 2147483647 on every row.
From the Trends MCP pull of 2026-09-03, a nike google search series held 261 Saturday rows. The first cell was 2021-09-04, index 71, volume 29,221,488. The last cell was 2026-08-29, index 64, volume 26,340,496. A get_growth call in the same session on that keyword printed metadata.total_data_points 261, recent_value 64.0, baseline_value 68.0 on 2025-08-30, growth -5.88%, volume_estimated true, and recent_volume 26,300,000. volume_growth was omitted with reason "volume is derived from the trend value for this source, not an independent measurement". Quote the 0-100 value on google search.
Cadence in this pull, by source:
| source | Points | First date | Last date | Step | Volume field |
|---|---|---|---|---|---|
| google search (nike) | 261 | 2021-09-04 | 2026-08-29 | weekly Saturday | yes, derived |
| google shopping (air fryer) | 260 | 2021-09-04 00:00:00+00:00 | 2026-08-22 00:00:00+00:00 | weekly | omitted |
| python (pandas) | 261 | 2021-08-30 | 2026-08-24 | weekly | yes, downloads |
| npm (react) | 261 | 2021-09-11 | 2026-09-05 | weekly | yes, downloads |
| amazon (air fryer) | 49 | 2022-07-31 | 2026-07-31 | month-end | yes |
| steam (Elden Ring) | 54 | 2022-03-01 | 2026-08-01 | month-start | yes, peak players |
| wikipedia (ChatGPT) | 45 | 2022-12-01 | 2026-08-01 | month-start | yes, page views |
| app downloads (com.whatsapp) | year-scale | 2025-09-06 | 2026-09-03 | irregular, often 2-day | yes, plus volume_cumulative |
npm react already had a 2026-09-05 cell (index 41.6, volume 72,793,284) while google search nike still ended 2026-08-29 and python pandas ended 2026-08-24. Amazon was a full month behind, last row 2026-07-31, index 51.5, volume 5,577,980. Wikipedia ChatGPT, returned as Chatgpt, peaked at 100.0 / 13,845 page views on 2025-07-01 and sat at 23.0 / 3,188 on 2026-08-01. Steam Elden Ring opened at 100.0 / 952,523 on 2022-03-01 and closed at 1.8 / 52,385 on 2026-08-01. For percent windows on these same series, see get_growth.
google search nike rows in this pull carried date, value, keyword, source, and volume. google shopping air fryer carried date, value, keyword, and source, with date in datetime form 2021-09-04 00:00:00+00:00 and no volume key at all. Last shopping index was 8.0 on 2026-08-22.
Amazon, Steam, Wikipedia, python, npm, and app downloads used datatype instead of source. Amazon air fryer rows looked like datatype: amazon. python pandas used datatype: python. A parser that only reads source will drop those series. A parser that assumes ISO dates without a time will break on google shopping.
Wikipedia also rewrote the keyword. The request used ChatGPT. Every returned row used Chatgpt. npm react and python pandas kept the requested lowercase names. Steam kept Elden Ring with the space and capitals. App downloads kept the Android bundle ID.
python pandas in this pull ran 261 weekly points from 2021-08-30 (index 2.5, volume 18,500,850) through 2026-08-24 (index 57.9, volume 121,976,673), with a 100.0 peak on 2026-08-03 at 200,739,859 downloads. npm react peaked at 100.0 on 2026-08-29 with 171,637,376 downloads, then fell to 41.6 on 2026-09-05. Those two developer sources share a weekly clock and a real volume field. They still do not share a latest date.
Steam Elden Ring printed index 0.0 on 2025-10-01 with volume 34,896. The 0-100 Steam index can floor while concurrent players remain in the tens of thousands. npm react printed index 0.0 on its first cell, 2021-09-11, with volume 2,343,676 downloads. Wikipedia Chatgpt printed 0.0 on 2022-12-01 with volume 1.
App downloads can print a true zero. com.openai.chatgpt on 2025-11-17 had value 0.0 and volume 0, with volume_cumulative held at 860,377,917 from the prior row. The same bundle ID later printed 100.0 on 2026-03-27 with volume 72,582,351. A chart that treats every 0.0 as missing data will erase real Steam and npm history and will also hide genuine empty app-download days.
The first get_time_series call on com.openai.chatgpt in this session returned HTTP 404 no_data. A retry on the same bundle ID returned the series from 2025-09-06 through 2026-09-01. A later get_growth call on that same pair returned no_data again. Treat app downloads as a source that can flap empty, then fill, then empty. The Android download series that did stay up, com.whatsapp, is the cleaner schema sample. Bundle ID rules for that source live on the Android app downloads API.
app downloads rows in this pull added volume_cumulative, a field google search, Amazon, Steam, Wikipedia, python, npm, and google shopping did not return. com.whatsapp printed volume_cumulative 2147483647 on every row from 2025-09-06 through 2026-09-03. That number is 2^31 minus 1, the signed 32-bit integer maximum. It did not move when period volume was 6,447,001 on the first row, 6,482,236 on 2026-09-03, or 0 on empty days such as 2026-02-23.
com.openai.chatgpt on the retry did increment. First row 2025-09-06: index 6.5, volume 4,729,735, cumulative 707,397,500. Last row 2026-09-01: index 6.2, volume 4,506,151, cumulative 1,516,810,080. The freeze is specific to a cumulative that has already hit the cap, not a missing default on the source.
WhatsApp also printed two rows for 2026-04-24: one with volume 7,348,562 and one with volume 0, both still capped at 2147483647 cumulative. Deduping on date is unsafe on this source. The latest WhatsApp cell was 2026-09-03, two days newer than the ChatGPT retry's 2026-09-01 close. Same source, two bundle IDs, two clocks.
A 2026-09-03 get_time_series body with source google search, amazon returned HTTP 400 invalid_source and message "'google search, amazon' is not a valid source." The same error for all said "'all' is not a valid source." Both payloads listed 16 valid values: google search, google images, google news, google shopping, youtube, wikipedia, tiktok, reddit, amazon, news sentiment, news volume, app downloads, app rankings, steam, npm, python. all is not on that list. Comma-separated lists belong on get_growth, not here.
Exact REST bodies from this pull:
{
"mode": "get_time_series",
"source": "google search",
"keyword": "nike"
}
{
"mode": "get_time_series",
"source": "amazon",
"keyword": "air fryer"
}
{
"mode": "get_time_series",
"source": "app downloads",
"keyword": "com.whatsapp"
}
MCP takes the same source and keyword fields. There is no data_mode, period, or country filter on the live tool schema. Product docs count each successful source plus keyword call as one request, including every point in the array. Failed 400 and 404 calls do not count. Free covers 100 requests per month. Starter is $19 for 1,000. Pro is $49 for 5,000. Business is $199 for 25,000. Annual billing saves 20%. Check current pricing before locking a budget.
Live ranked boards are a different tool. get_time_series will not return Steam Most Played or Google Trends leaders. Those feeds, with limit default 25 and max 200, sit on get_top_trends pagination. Custom {recent, baseline} windows are REST-only on get_growth.
FAQ