Hugging Face Hub Scraper - Models, Datasets & Spaces

Extract public metadata for Hugging Face models, datasets and Spaces in one run — likes, downloads, trending score, pipeline task, license, tags, created/updated dates. Built on the official Hub JSON API. No login, no browser.

Pay-per-event pricing (when enabled by the owner): $1.50 / 1,000 results (result event) + default Actor start ($0.00005). Platform usage is not charged on top when PPE is configured without "+ usage".

Why this Actor

Capability	Included
Models + datasets + Spaces, unified schema	✅
Sort by trending / likes / downloads / date	✅
Filter by author, keyword, tags	✅
Only-new-since-last-run (schedule-friendly)	✅
Official Hub API, polite rate limit + 429 retry	✅

Use cases

AI trend monitoring · open-model selection · license compliance screening · newsletter / research feeds · investment research.

Input

Field	Description
repoTypes	models, datasets, spaces (any combination). Prefill scrapes all three.
search / author / filterTags	Optional filters (e.g. llama, google, text-generation, license:apache-2.0)
sort	trendingScore (default), likes, downloads, lastModified, createdAt
maxItemsPerType	Cap per type (default 50; paginates 100/page via Link header)
onlyNew	Remember seen IDs in KV store hf-hub-state and return only new repos

Example:

{ "repoTypes": ["models", "datasets", "spaces"], "sort": "trendingScore", "maxItemsPerType": 50 }

Output

One dataset row per repo: repoType, id, author, url, likes, downloads, trendingScore, pipelineTag, libraryName, sdk, license, gated, tags, createdAt, lastModified, scrapedAt. Views: Overview, Trending leaderboard, License screening, All fields. Run counters and warnings: key-value record RUN_STATS. See sample-output.json.

Pricing notes

• Event result: one Hub repo (model / dataset / Space) with public metadata — $0.0015 each ($1.50 / 1,000).
• Event apify-actor-start: default start charge.
• Respects the run's Max cost per run — the Actor stops when the budget is reached.
• Do not enable both result and apify-default-dataset-item (would double-charge).

Use via Apify MCP (Claude, Cursor, Windsurf, Copilot)

This Actor is a normal Store Actor with pay-per-event pricing, so it works with Apify's hosted MCP server and with agentic payments (prepaid tokens / Skyfire-style pay-per-request). Monetization stays on Apify — no Smithery listing required.

Pin this Actor as a tool (recommended for agents):

https://mcp.apify.com?tools=infinite_seascape/hf-hub-trending-scraper

Or let the agent discover it with the default search-actors / call-actor tools at https://mcp.apify.com.

Example agent input (keep maxItemsPerType ≤ 20 — call-actor waits ~30s by default):

{ "repoTypes": ["models"], "sort": "trendingScore", "maxItemsPerType": 10, "filterTags": ["text-generation"] }

After call-actor returns, read rows with get-dataset-items using the run's datasetId. Cap spend with callOptions.maxTotalChargeUsd (not an Actor input field).

Docs: https://docs.apify.com/mcp · Configurator: https://mcp.apify.com/

Run locally

npm install
# edit storage/key_value_stores/default/INPUT.json
npm start   # or: apify run

Compliance

Uses the documented public Hub API (/api/models, /api/datasets, /api/spaces). huggingface.co/robots.txt allows /. 500 ms delay between pages; retry/backoff on 429. No personal data beyond public usernames/org names. Review https://huggingface.co/terms-of-service before commercial use. Not affiliated with Hugging Face.

Tutorial: build a daily AI trend radar from Hugging Face in 10 minutes

Goal: every morning, get the Hugging Face models, datasets and Spaces that started trending since yesterday — as JSON/CSV, a Google Sheet, or a Slack message.

Actor: Hugging Face Hub Scraper - Models, Datasets & Spaces (infinite_seascape/hf-hub-trending-scraper). Pricing: $1.50 per 1,000 results (+ $0.00005 per run start).

1. First run: see what's trending

Input:

{ "repoTypes": ["models", "datasets", "spaces"], "sort": "trendingScore", "maxItemsPerType": 20 }

You get 60 rows (20 per type) with repoType, id, likes, downloads, trendingScore, pipelineTag, license, tags, lastModified, url. Open the Overview view in the Output tab.

Cost: 60 results × $0.0015 ≈ $0.09.

2. Narrow it down

• Only text-generation models with a permissive license:

{ "repoTypes": ["models"], "filterTags": ["text-generation", "license:apache-2.0"], "sort": "trendingScore", "maxItemsPerType": 50 }

• One organisation: "author": "google" (a full https://huggingface.co/google URL also works).
• Keyword: "search": "embedding". Tip: the Hub API is sometimes slow for trending + search; the Actor automatically retries sorted by likes and tells you in the log.

3. Turn on "only new"

{ "repoTypes": ["models", "datasets", "spaces"], "sort": "trendingScore", "maxItemsPerType": 50, "onlyNew": true }

The Actor remembers every repo it has returned (key-value store hf-hub-state) and on the next run outputs only repos it hasn't seen. You pay only for new results.

4. Schedule it

Console → Schedules → Create → cron 0 7 * * * (daily 07:00) → add this Actor with the input above.

5. Send results somewhere

Actor → Integrations: Slack, Google Sheets, Zapier, Make, n8n or a webhook on Run succeeded. Or call the API:

curl "https://api.apify.com/v2/acts/infinite_seascape~hf-hub-trending-scraper/run-sync-get-dataset-items?token=<YOUR_TOKEN>" \
  -H 'Content-Type: application/json' -d '{"repoTypes":["models"],"maxItemsPerType":10}'

6. Keep costs predictable

Set Max cost per run (e.g. $0.20). The Actor stops cleanly when the budget is reached and says so in the status message.

Good to know

• Data comes from the official public Hub API; no login, no private repos.
• downloads is the Hub's recent-downloads figure (not shown for Spaces).
• Not affiliated with Hugging Face.