GitHub Trending Repositories Search

Purpose

Return the current set of trending public repositories from github.com/trending, with optional filters for spoken language (e.g. Chinese), programming language (e.g. Python), and date range (daily / weekly / monthly). Each entry includes owner/repo name, description, programming language (+ language color), total stars, total forks, stars gained in the selected period, and the top contributing developers ("Built by"). Read-only — no auth, no cookies, no state mutation.

When to Use

• Daily / weekly / monthly monitoring of what is currently trending on GitHub.
• Tracking trending repos within a specific programming language (e.g. "trending Rust this week").
• Tracking trending repos by spoken language (e.g. trending repos whose README is in Chinese, Japanese, Spanish — GitHub detects this from the README's natural-language content, not from anything the maintainer declares).
• Anywhere you'd want a server-rendered, parseable list of "what's hot on GitHub right now" — there is no official GitHub REST/GraphQL endpoint for this; the /trending HTML page is the canonical surface.

Workflow

GitHub Trending is a plain server-rendered HTML page with no auth, no anti-bot, no JS rendering, and no XHR/GraphQL data layer behind it. The fastest, cheapest, most reliable path is to GET the URL with a vanilla User-Agent and parse the HTML — one HTTP request, ~1 second wall, no LLM cost. Always lead with the fetch path. A scripted browser session works as a fallback (the page snapshots cleanly via browse snapshot — verified at 558 a11y refs in a 16-repo run), but it pays a ~30× wall-clock penalty and burns inner-agent turns for zero additional information vs. the raw HTML.

1. Build the URL

The three filters compose like this:

Full example (all three filters): https://github.com/trending/python?since=daily&spoken_language_code=zh

The unfiltered base URL https://github.com/trending works too — it defaults to since=daily and "all languages."

2. Fetch the HTML

# via the browse CLI (no proxies needed, no stealth needed)
browse cloud fetch "https://github.com/trending/python?since=daily&spoken_language_code=zh"

# or any HTTP client — a vanilla User-Agent + Accept: text/html is fine
curl -sS -H 'User-Agent: Mozilla/5.0' \
  'https://github.com/trending/python?since=daily&spoken_language_code=zh'

Returns 200 OK with Content-Type: text/html; charset=utf-8 for anonymous unauthenticated requests. No residential proxy required, no --verified, no session, no cookies — verified 2026-05-21 from a Vercel Sandbox IP, 0 failed requests across 214 page requests in browser trace.

3. Parse the HTML

Each repo is wrapped in <article class="Box-row">. Iterate articles, then extract within each one:

4. Validate filter application

GitHub renders filter chips in the toolbar that confirm which filters were applied. The most reliable verification is to inspect the page title in the response: <title>Trending {Lang} repositories on GitHub today · GitHub</title> (daily), … this week · GitHub (weekly), … this month · GitHub (monthly). The title omits "spoken language" but reflects programming language + period.

5. Browser fallback

When direct HTTP is blocked (network egress restrictions, captive portals, corporate proxies that MITM github.com), drive a Browserbase session:

SID=$(browse cloud sessions create --keep-alive | node -e "let s='';process.stdin.on('data',c=>s+=c).on('end',()=>process.stdout.write(JSON.parse(s).id))")
browse open "https://github.com/trending/python?since=daily&spoken_language_code=zh" --remote --session "$SID"
browse wait load --remote --session "$SID"
browse snapshot --remote --session "$SID"  # ~558 a11y refs — every article exposed as headings + links
browse cloud sessions update "$SID" --status REQUEST_RELEASE

--verified and --proxies are not needed (verified 2026-05-21 with a bare session — 0 anti-bot blocks across full trace). Snapshot gives you all 16 article cards directly; alternatively browse get html body returns the same HTML the fetch path returns and you parse with the same selectors.

Site-Specific Gotchas

• No official API for Trending. GitHub's REST and GraphQL APIs both have search/repositories endpoints that can sort by stars over a date range, but neither reproduces the editorial signal that powers /trending (which factors in star velocity, dedupes recently-popular repos, etc.). Don't waste time trying to back-derive /trending from GET /search/repositories — the result sets diverge significantly. The HTML page is canonical.
• Accept: application/json is ignored. /trending only renders HTML. No JSON variant, no RSS feed, no Atom feed.
• Programming language is a URL path segment, not a query param. ?language=python is silently ignored — the page renders the all-language list. Use /trending/python instead. Slugs are lowercase; multi-word languages use the hyphenated form GitHub displays in the "Language" dropdown (e.g. objective-c, common-lisp, vim-script). For C++ use c%2B%2B (URL-encoded + is required; c++ works in browsers because they percent-encode automatically but raw HTTP clients won't).
• Spoken language code is ISO 639-1, two-letter. zh (Chinese), ja (Japanese), es (Spanish), pt (Portuguese), ko (Korean), ru (Russian), de, fr, etc. The dropdown surfaces all valid codes in the page's <option value="spoken_language_code=XX"> elements — there is no separate API enum. Invalid codes are silently dropped and the page renders the all-languages list. Combining spoken_language_code with a programming-language path segment is allowed and works as AND.
• stars_period text uses singular "star" when the count is 1. /(\d+)\s*stars\s+today/ will miss "1 star today" rows. Use stars? in your regex.
• stars_period can be missing entirely — not all rows have the <span class="d-inline-block float-sm-right">N star(s) period</span>. Observed 1 of 16 missing in a single fetch (~6%). Treat the field as nullable; do not default to 0 (the absence likely means the period delta is small/zero/not-yet-computed by GitHub, but the page is unambiguous about not displaying anything).
• The "Built by" list is top-5 period contributors, not the repo's owners. Maintainers, dependabot, github-actions, claude, and similar bot accounts routinely show up here. Don't conflate built_by with owner.
• Heading anchor extraction must be scoped to the <h2>. A naïve <a href="/{owner}/{repo}"> regex over the whole article will match the first href in the article — which is the "log in to star" affordance <a href="/login?return_to=%2F{owner}%2F{repo}"> followed by <a href="/{owner}/{repo}/stargazers">. Both produce wrong values (the login href URL-encodes the slashes; the stargazers href produces repo: "actualrepo/stargazers"). Match inside <h2 class="h3 lh-condensed"> … </h2> to get the canonical repo link.
• Description class name has two observed variants. Currently col-9 color-fg-muted my-1 tmp-pr-4 (post-Primer-CSS migration); historically and still seen in cached responses as col-9 color-fg-muted my-1 pr-4. Match (?:tmp-)?pr-4.
• Max ~25 repos per page; no pagination. GitHub intentionally caps Trending to a single page (typical: 15–25 entries depending on filter density). There is no ?page=2. If you need more, change the date range to weekly or monthly — those produce different (broader) result sets.
• No rate-limit observed for anonymous fetches at ≤ 1 req/s. GitHub's standard anonymous rate limit (60 req/hr) applies to the API; HTML pages are throttled separately and /trending returns 200 reliably at human-paced polling. Authenticated requests count against the 5000 req/hr API budget but Trending is fine unauthenticated.
• Third-party "trending API" mirrors are unreliable. ghapi.huchen.dev, trendings.herokuapp.com, and similar Heroku/Cloudflare-Workers mirrors of the page exist but go offline regularly (their HTML scrapers break when GitHub re-themes the page — and several of the well-known ones haven't been updated since the 2024 Primer-CSS migration). Always go to source.

Expected Output

{
  "success": true,
  "url": "https://github.com/trending/python?since=daily&spoken_language_code=zh",
  "filters": {
    "programming_language": "python",
    "spoken_language_code": "zh",
    "since": "daily"
  },
  "count": 16,
  "repos": [
    {
      "rank": 1,
      "owner": "hiroi-sora",
      "repo": "Umi-OCR",
      "url": "https://github.com/hiroi-sora/Umi-OCR",
      "description": "OCR software, free and offline. 开源、免费的离线OCR软件。支持截屏/批量导入图片，PDF文档识别，排除水印/页眉页脚，扫描/生成二维码。内置多国语言库。",
      "language": "Python",
      "language_color": "#3572A5",
      "stars_total": 44362,
      "forks_total": 4379,
      "stars_period": 38,
      "period_label": "stars today",
      "built_by": ["hiroi-sora", "weblate", "chunkiuu", "qwedc001", "plum7x"]
    },
    {
      "rank": 12,
      "owner": "xrayfree",
      "repo": "free-ssr-ss-v2ray-vpn-clash",
      "url": "https://github.com/xrayfree/free-ssr-ss-v2ray-vpn-clash",
      "description": "长期免费维护 VLESS/ VMess / Trojan / SS / V2RAY / VPN / CLASH / 小火箭 免费节点订阅链接！电报群：https://t.me/xrayfree",
      "language": "Python",
      "language_color": "#3572A5",
      "stars_total": 1594,
      "forks_total": 75,
      "stars_period": 1,
      "period_label": "stars today",
      "built_by": ["xrayfree", "github-actions"]
    }
  ],
  "error_reasoning": null
}

Failure shape (only realistic outcome — github.com/trending is operationally extremely reliable; the most likely failure is your egress being unable to reach github.com at all):

{
  "success": false,
  "url": "https://github.com/trending/python?since=daily&spoken_language_code=zh",
  "filters": { "programming_language": "python", "spoken_language_code": "zh", "since": "daily" },
  "count": 0,
  "repos": [],
  "error_reasoning": "fetch failed: ECONNREFUSED github.com:443"
}

Field types:

• count — integer; equals repos.length. Typically 15–25.
• repos[].description / repos[].language / repos[].language_color / repos[].stars_period / repos[].period_label — all nullable. Treat missing-in-DOM as null, not 0 / "".
• repos[].stars_total / repos[].forks_total — integers, parsed from comma-formatted display strings.
• repos[].built_by — array of GitHub usernames (strings), may be empty.
• period_label — one of "stars today", "stars this week", "stars this month", or null. The label tracks the since filter; expect consistency within a single result page.