Pass source google search for a weekly Google Search interest series, or source python for a weekly PyPI download series. python volume is independent. Search volume is often missing. PyPI project names are not import paths.
Live data as of 2026-10-14
Source google search on get_growth and get_time_series returns a weekly 0-100 Google Search interest series for one phrase. Source python returns a weekly 0-100 download-index series for an exact PyPI project name, plus an independent download-count volume field. Both pulls on October 14, 2026 used the string pandas. Google Search held 261 weekly points from 2021-10-16 through 2026-10-10 and closed 28.0, with no volume keys. python held 260 weekly points from 2021-10-11 through 2026-09-28 and closed 67.8, with series volume 140569797. Search 12M fell 45.1 percent from 51.0. python 12M rose 39.22 percent from 48.7, while download volume rose 34.14 percent from 104795457.0. The live boards use type Google Trends and type GitHub, not those source strings.
The October 14, 2026 pull is the split that still matters next month: weekly Search interest versus weekly PyPI downloads, then two indexes that do not share a scale. get_time_series for pandas on source google search returned 261 weekly points from 2021-10-16 (value 55, no volume field) through 2026-10-10 (value 28). Peak 100 sat on 2022-04-09. The series field is source google search. Growth metadata printed total_data_points 261. Search growth rows printed no volume_available, no volume_estimated, and no recent_volume.
get_time_series for pandas on source python returned 260 weekly points from 2021-10-11 (value 3.0, volume 19314154) through 2026-09-28 (value 67.8, volume 140569797). Peak 100.0 sat on 2026-08-03 at volume 200739859. The series field is datatype python, not source. REST and MCP both printed datatype. Growth metadata printed total_data_points 260. python starts five days earlier. Search does not backfill that first python week. The python close sits twelve days behind the Search close. Quote both dates. The lag is the series, not a stale live board.
Search 7D matched 1W exactly: minus 24.32 percent from 37.0, recent_date 2026-10-10, baseline 2026-10-03. python 7D matched 1W exactly: plus 8.48 percent from 62.5, recent_date 2026-09-28, with volume_growth plus 7.66 percent from 130562973.0. 30D matched 1M on both. Search 30D was minus 12.5 percent from 32.0. python 30D was plus 17.1 percent from 57.9, download volume plus 15.14 percent from 122089427.0. Search YTD was minus 9.68 percent from 31.0 on 2026-01-03. python YTD was plus 62.2 percent from 41.8 on 2025-12-29, with volume_growth plus 52.89 percent from 91944452.0. Omitting percent_growth defaults to 12M on both and succeeded on both.
| Keyword | Search close | Search 12M | Search YTD | python close | python 12M index | python 12M downloads | Points |
|---|---|---|---|---|---|---|---|
| pandas | 28.0 on 2026-10-10 | -45.1% from 51.0 | -9.68% from 31.0 | 67.8 on 2026-09-28 | +39.22% from 48.7 | +34.14% (140569797 from 104795457) | 261 / 260 |
| requests | 28.0 on 2026-10-10 | -42.86% from 49.0 | -22.22% from 36.0 | 66.7 on 2026-09-28 | +56.21% from 42.7 | +49.41% (304039543 from 203492184) | 261 / 260 |
| numpy | 24.0 on 2026-10-10 | -58.62% from 58.0 | 0.0% from 24.0 | 67.2 on 2026-09-28 | +75.46% from 38.3 | +65.47% (207208807 from 125224131) | 261 / 260 |
| pydantic | 14.0 on 2026-10-10 | -73.08% from 52.0 | -30.0% from 20.0 | 73.6 on 2026-09-28 | +93.18% from 38.1 | +90.94% (213434337 from 111780259) | 261 / 260 |
| fastapi | 11.0 on 2026-10-10 | -72.5% from 40.0 | -42.11% from 19.0 | 67.8 on 2026-09-28 | +179.01% from 24.3 | +175.33% (102326633 from 37164665) | 261 / 260 |
The 12M sign flip is not a pandas quirk. requests, numpy, pydantic, and fastapi each fell on Search and rose on python over the same 12M window. langchain did the same at 12M (Search minus 76.81 percent from 69.0, python plus 116.88 percent from 23.7) on 261 Search points versus 206 python points. openai Search 12M was minus 83.08 percent from 65.0, while python 12M was plus 156.97 percent from 25.1 and download volume plus 156.98 percent (73950035.0 from 28776125.0). Quote both closes. Do not rescale one index onto the other.
python can print index 0.0 while downloads remain. pandas on 2021-12-27 printed value 0.0 and volume 13792668. A zero index is not an empty registry week. Read volume before treating 0.0 as missing data. Search volume is estimated when present, and it was absent here. pandas, requests, numpy, pydantic, fastapi, langchain, and openai printed no Search volume keys on this pull. Presence of a volume key on one Search keyword, as react printed on source npm in a sibling pull, does not promise it on the next.
Source names case-fold. Python and PYTHON on get_growth for pandas returned the same 12M row as the lowercase label. Feed labels do not belong here. Source GitHub returned HTTP 400 invalid_source, message 'github' is not a valid source., and listed 16 keyword sources plus all. Source Google Trends did the same, echoed as 'google trends'. The handler lowercased the rejected type string. It did not map it to source python or source google search.
Google Search pandas printed no volume. python pandas printed independent download counts. A comma call still mixes the first windows, then weights only the source that has volume.
{"mode": "get_growth", "source": "google search, python", "keyword": "pandas", "percent_growth": ["7D", "30D", "12M", "YTD"]}
That comma call on October 14, 2026 returned sources_successful 2. trend_analysis printed uptrend_count 1, downtrend_count 1, and uptrend_ratio 1/2. Both 7D windows succeeded. python 7D was plus 8.48 percent. Search 7D was minus 24.32 percent. The 1/2 tally used those first windows. Search 12M minus 45.1 percent never entered that ratio as a second vote. python 12M plus 39.22 percent never entered either. The mixer did see a sign split this time, at 7D, because the two 7D windows already disagreed. That is the opposite of the react Search-versus-npm mix, where both 7D windows were down and uptrend_ratio printed 0/2.
volume_weighted_growth was plus 7.66 with has_volume_data true, sources_with_volume 1, weight_coverage 1/2 sources, and total_volume 140569797.0. python contributed 140569797.0 (100.0 percent of the mix) at growth plus 7.66, weighted contribution plus 7.66. Google Search contributed nothing to that mix. The plus 7.66 is python volume_growth, an independent download-count change. Search printed no recent_volume and no volume_growth. The mixer did not invent Search units to fill the gap. Read source_results before treating 7.66 as a blended Search-plus-PyPI number. Extra windows still sit on one get-growth request per source.
get_time_series rejects the comma. Source google search, python returned HTTP 400 invalid_source and listed 16 keyword sources with no all. One source per series call. Two sources belong on get-growth.
Source google search accepts an ordinary phrase. Source python is documented as an exact PyPI project name. Live traffic enforced the second rule more strictly than case.
Pandas and PANDAS folded on both sources. python printed the same 67.8 / 140569797.0 close and the same 12M plus 39.22 percent. Search printed the same 28.0 close and the same 12M minus 45.1 percent. python keywords case-fold. That differs from source npm, where React returned no_data. Prefer the canonical lowercase PyPI name in stored configs anyway.
Import aliases miss. pd on python closed 17.2 with 2393.0 downloads, 12M minus 75.07 percent from 69.0. That is a different project, not the pandas alias people type in notebooks. Search pd closed 59.0, 12M minus 15.71 percent from 70.0, an English abbreviation series. pandas.py on python returned HTTP 200 with envelope statusCode 502 source_unavailable. Search pandas.py closed 0.0 with 12M 0.0 percent from 0.0 and attached data_quality warnings. Treat 502 on python as "no series" until a later pull returns a row.
The sklearn trap is the one that still bites next month. sklearn on python closed 2.3 with 362985.0 downloads. 12M was minus 86.06 percent from 16.5, volume minus 37.58 percent from 581528.0. 7D index plus 91.67 percent from 1.2 sat next to volume plus 5.22 percent. 30D index plus 475.0 percent from 0.4 sat next to volume plus 8.75 percent. Tiny indexes swing hard. scikit-learn is the project people install: close 58.5, volume 44392259.0, 12M plus 46.62 percent from 39.9, download volume plus 39.83 percent from 31747650.0. Search sklearn closed 15.0, 12M minus 64.29 percent from 42.0. Search scikit-learn closed 1.0, 12M minus 92.86 percent from 14.0. The hyphenated project name barely registers as a Search phrase. The import name barely registers as a PyPI install. Join on the pypi.org slug.
The 0-100 columns are independent indexes. pandas Search close 28.0 is not 28 percent of python close 67.8. python 12M plus 39.22 percent from 48.7 is a jump on the download index, not a matching rise in Search demand. Search 12M on that name was minus 45.1 percent from 51.0. Quote both closes. Do not add the percents. Weekly PyPI follow-up lives on the PyPI package downloads API. Google Search history lives on google search data.
Two strings, two tools. REST uses mode get_growth or get_top_trends. MCP uses the matching tool names. The adjacent live board is python vs GitHub, not a second python keyword source.
{"mode": "get_growth", "source": "google search, python", "keyword": "pandas", "percent_growth": ["30D", "12M"]}
{"mode": "get_time_series", "source": "google search", "keyword": "pandas"}
{"mode": "get_time_series", "source": "python", "keyword": "pandas"}
{"mode": "get_top_trends", "type": "Google Trends"}
{"mode": "get_top_trends", "type": "GitHub"}
The MCP equivalents are get_growth(keyword="pandas", source="google search, python", percent_growth=["30D", "12M"]), get_time_series(keyword="pandas", source="google search"), get_time_series(keyword="pandas", source="python"), get_top_trends(type="Google Trends"), and get_top_trends(type="GitHub"). REST 1M is not required here. Both sources are weekly, so 7D succeeded on both for pandas.
Omitting limit is not required to learn the stamp. Type Google Trends with limit 1 returned as_of_ts 2026-10-14T06:01:49.638247+00:00, rank 1 rj mitte. Type GitHub with limit 1 returned the same stamp, rank 1 morluto/rea. Rank is live chart order, not a sort of the 2026-09-28 python index. Rank 1 on that board is a repository name. It is not a PyPI project keyword.
REST type google search (the keyword source string, lowercase) returned HTTP 200, echoed type Google Trends, and matched rj mitte at rank 1. REST type Google Search did the same. REST type python and type Python returned HTTP 400, error invalid_request, message Unknown type 'python' (or 'Python'), plus 24 valid feed labels. REST type steam, type youtube, and type npm returned the same 400 shape. type youtube trending aliased to YouTube Trending. llms.txt still describes feed labels as case-sensitive. The live handler accepted some lowercase aliases on October 14, 2026 and rejected others. Prefer the canonical spelling in stored configs. Do not pass source python as a feed type.
REST omitting type is a different 400. POST https://api.trendsmcp.ai/api with mode get_top_trends and no type returned HTTP 200 with envelope statusCode 400, error invalid_request, message The 'type' parameter is required, then the same 24-name list. llms.txt still names that miss missing_parameter. Read statusCode inside the JSON, not only the HTTP line.
get_top_trends MCP pricing is 1 credit per 10 requested rows, max 10 credits per call. REST counts 1 request per call. get_growth bills one request per source plus keyword, with extra percent_growth windows included, so a two-source comma string is two units. get_time_series bills one request per source plus keyword. Keyword follow-up stays on source google search or source python. The live boards stay on their type strings.
FAQ