"YouTube transcripts for RAG without getting blocked on cloud servers"

Nathan Carter — Mon, 05 Oct 2026 11:33:01 +0000

If you've tried fetching YouTube transcripts from a server, you've probably seen it: the same code that works on your laptop fails on AWS, GCP or Vercel. YouTube asks data-centre IPs to "sign in to confirm you're not a bot", and its web player now needs a proof-of-origin token for captions.

I needed transcripts for a RAG pipeline, so I built a small hosted scraper that handles this: YouTube Transcript Scraper on Apify. It reads captions through YouTube's app endpoint and retries on residential IPs when a server IP is challenged.

Three lines of Python

from apify_client import ApifyClient

client = ApifyClient("YOUR_APIFY_TOKEN")
run = client.actor("automationnation/youtube-transcript-scraper").call(run_input={
    "videos": ["https://www.youtube.com/watch?v=UF8uR6Z6KLc", "https://www.youtube.com/@3blue1brown"],
    "maxVideosPerChannel": 5,
})
docs = list(client.dataset(run["defaultDatasetId"]).iterate_items())

Each row has:

transcript: the full text
segments: [{"start": 6.05, "duration": 1.0, "text": "..."}] for citing timestamps
title, channel, viewCount, durationSeconds, description, keywords
transcriptLanguage, isAutoGenerated and every caption language the video has
status: ok, no_transcript or unavailable (only ok rows are charged)

Into a vector store

chunks = []
for d in docs:
    if d["status"] != "ok":
        continue
    window, start = [], d["segments"][0]["start"]
    for s in d["segments"]:
        window.append(s["text"])
        if len(" ".join(window)) > 800:
            chunks.append({"text": " ".join(window), "url": f'{d["url"]}&t={int(start)}s', "title": d["title"]})
            window, start = [], s["start"]

Every chunk links back to the exact second of the video, so your answers can cite timestamps.

Cost

$1.50 per 1,000 transcripts, and videos without captions are free. Apify's free tier ($5 a month) covers about 3,300 transcripts.

From an AI agent

Add https://mcp.apify.com/?tools=automationnation/youtube-transcript-scraper to Claude or Cursor and ask "Summarise the last five videos on @lexfridman with timestamps."

"Google Shopping, Flights, Hotels and News as JSON in Python: 8 pay-per-result scrapers"

Nathan Carter — Mon, 05 Oct 2026 11:32:46 +0000

Failed searches aren't charged, and you can cap the cost of any run. All of them are on the docs site with more examples.