GitHub Trending rank_change climbers via get_top_trends

GitHub rank_change on Trends MCP is get_top_trends with type GitHub, sort rank_change, and a window such as 1d. From the August 27, 2026 pull, the board stamped 2026-08-27T14:01:41 UTC. The 1d window used baseline 2026-08-27T02:01:32 UTC, a snapshot from the same calendar day, and returned count 4. K-Dense-AI/scientific-agent-skills led that list at rank 6, prev_rank 14, rank_change 8. Default sort rank closed at 18 owner/name rows, then count 0 at offset 18. Rank 1 was zedeus/nitter. Wikipedia Trending 1d on the same stamp returned count 42. Free tier access is 100 requests per month.

GitHub 1d returned 4 climbers on a same-day 02:01 baseline

From the Trends MCP scan of August 27, 2026, GitHub sort rank_change stamped 2026-08-27T14:01:41 UTC and filled 4 climber rows on window 1d against baseline 2026-08-27T02:01:32 UTC, so a 1d GitHub pull can diff two scans on the same date rather than yesterday versus today.

Default sort rank returned 18 repos at offset 0 with limit 50, then count 0 at offset 18. Rank 1 was zedeus/nitter. Rank 2 freestylefly/awesome-gpt-image-2. Rank 3 tt-a1i/archify. Rank 4 JetBrains/go-modern-guidelines. Rank 5 anthropics/claude-plugins-official. Rank 12 google/googletest. Rank 16 actions/checkout. Rank 17 OpenCut-app/OpenCut. zedeus/nitter sat at rank 1 and was absent from the 4-row climber list. Passing type GitHub Trending Repos on the same 1d call still returned "type": "GitHub" with the same 4 objects, so treat GitHub as the canonical label.

Same-stamp Wikipedia Trending 1d used that 02:01 baseline and returned count 42, led by 2026-27 UEFA Europa League at rank 78, prev_rank 179, rank_change 101. Google Trends 1d returned count 22, led by hurricane dolly at rank 75, prev_rank 156, rank_change 81. App Store Top Free 1d returned count 45, led by Netflix at rank 58, prev_rank 90, rank_change 32. Google News Top News 1d returned count 1: the Ottumwa Pizza Ranch measles headline at rank 37, prev_rank 38, rank_change 1. ChatGPT sat at App Store Top Free rank 1 on this stamp and was absent from GitHub 18.

The GitHub Trending Repos API page still describes a 12-row scan from August 1, 2026 led by zhaoxuya520/reverse-skill. This pull closed at 18, with zedeus/nitter at rank 1.

Exact API calls for climber boards

sort rank_change plus window is a separate get_top_trends request from default sort rank, so a 1d climber pull and a current-rank pull on August 27, 2026 are two requests, and each extra window is another request because window is per call.

{
  "mode": "get_top_trends",
  "type": "GitHub",
  "sort": "rank_change",
  "window": "1d",
  "limit": 50
}

{
  "mode": "get_top_trends",
  "type": "GitHub",
  "sort": "rank_change",
  "window": "7d",
  "limit": 50
}

{
  "mode": "get_top_trends",
  "type": "GitHub",
  "limit": 50,
  "offset": 0
}

MCP clients send the same fields on get_top_trends. Window values are 1d, 3d, 7d, 14d, and 30d. Default sort is rank. Default window when sort is rank_change is 30d, so omit window only when that 30d lookback is the intent. On this stamp the 30d call returned count 0 with a complete-turnover note, which is a paid request that still answers a real question: none of the current 18 repos sat on the July 29 snapshot. Offset and limit behavior is on get_top_trends pagination. For the Google Trends sibling, see Google Trends rank_change. For a current-rank join against Google Trends, see GitHub Trending vs Google Trends.

Scrapers of github.com/trending return a current page. They do not emit prev_rank against a stored Trends MCP snapshot. Those products are not substitutes. Check current pricing on any scrape vendor before quoting a fee.

3d held or fell while 7d turned over completely

A longer rank_change window on GitHub can return fewer rows, not more, because only repos still on the current 18-row board that also sat on the older snapshot can climb, and the note text changes meaning when count is 0.

Every window on this stamp used the 02:01 cluster. 1d compared 14:01 against 02:01 on August 27, about twelve hours, not a full calendar day. 3d found overlap with the August 24 board and still returned no climbers, which is why the note says everything held or fell. 7d, 14d, and 30d found zero shared repos, so rank_change cannot be computed. Count 0 is not one error. Read the note before retrying a shorter window or switching to sort rank.

K-Dense-AI/scientific-agent-skills led 1d at +8 and had no 3d row because that window returned an empty list. DietrichGebert/ponytail, ConardLi/garden-skills, and marin-community/marin only appear as climbers on the 1d call. Rank 1 zedeus/nitter never entered those objects. A current leader can sit outside the climber list when it held, fell, or was missing from the baseline snapshot.

Live GitHub ranks and search indexes can disagree

There is no github source on get_growth. Star history is not in this payload. Follow-ups use google search or youtube on a related keyword, each billed as its own request, and a high GitHub rank can sit next to a flat or falling search index.

On the 2026-08-22 google search points, nitter sat at 33.0, up 3.12% over 7 days from 32.0 and down 23.26% over 30 days from 43.0. youtube for nitter sat at 27.0, up 145.45% over 7 days from 11.0 and down 65.38% over 30 days from 78.0. google/googletest sat at GitHub rank 12 while googletest on google search sat at 14.0, down 39.13% over 7 days from 23.0 and down 76.27% over 30 days from 59.0. OpenCut-app/OpenCut sat at GitHub rank 17 while opencut on google search sat at 36.0, down 7.69% over 7 days from 39.0. The 1d top climber, scientific agent skills, sat at google search 10.0, up 100.0% over 7 days from 5.0 and down 75.61% over 30 days from 41.0. Quote both values when the baseline is small. A doubling off 5.0 is not the same story as a doubling off 50.0.

Unofficial GitHub trending scrapers still advertise daily, weekly, and monthly HTML copies. Trends MCP rank_change is a snapshot diff with prev_rank, not a stars-since field from that HTML. The August 27, 2026 1d list had four rows. Publish those four. Do not pad to 25.