Collectors · Social & communities
YouTube transcript scraper in Python
Transcripts of YouTube videos in bulk: full text, timestamped segments, language and chapters. One POST from Python with videos, rows back as JSON with 17 fields, $0.002 per delivered video. No proxies or parsers to maintain.
First call
Install, call, read the rows.
pip install requestsimport os, requests
r = requests.post(
"https://api.quanticdata.io/v1/scraper/collectors/youtube_transcript/run",
headers={"Authorization": f"Bearer {os.environ['QD_API_KEY']}"},
json={
"videos": ["dQw4w9WgXcQ"],
"max_results": 1,
},
timeout=120,
)
r.raise_for_status()
run = r.json()["payload"]
print(run["count"], "rows,", run["cost"]["usd"], "USD")
for row in run["results"]:
print(row.get("video_id"), row.get("url"), row.get("title"), row.get("channel"))The key comes from an environment variable, never from the code. The run answers in the same request when it finishes in time; payload.results holds the rows and payload.cost what they cost.
Request parameters
What the collector takes.
| Parameter | Type | Required | Meaning |
|---|---|---|---|
videos | string[] | yes | Video ids, or watch, shorts and youtu.be URLs, one per line, up to 50. |
languages | string[] | no | Caption languages to prefer, in order (e.g. en, it). Omit for the spoken language. |
include_without_captions | boolean | no (default false) | Deliver (and bill) a row with has_transcript=false for videos that have no captions (default false). |
max_segments | integer | no (default 2000) | Timestamped segments per row (1–5000, default 2000). text always holds the whole transcript. |
max_results | integer | no (default 10) | How many videos to deliver at most (1–50). You pay only for delivered videos. |
Response fields
17 fields per video.
| Field | Type | What it holds |
|---|---|---|
video_id | string | 11-char video id. |
url | string | Watch URL. |
title | string | Video title. |
channel | string, nullable | Channel name. |
channel_id | string, nullable | Channel id (UC…). |
duration_seconds | integer, nullable | Length in seconds. |
has_transcript | boolean | True when a transcript was read. |
transcript_source | string, nullable | manual or automatic captions. |
language | string, nullable | Language of the transcript. |
caption_languages | string[] | Caption languages the video offers. |
text | string, nullable | Whole transcript as plain text. |
word_count | integer | Words in text. |
segment_count | integer | Timestamped segments in the transcript. |
segments | object[] | Timestamped segments: start and duration in seconds, text. |
segments_truncated | boolean | True when max_segments cut the segment list (text stays whole). |
chapters | object[] | Author's chapters: start in seconds, title. |
rank | integer | 1-based delivery order. |
Add ?format=csv to the runs endpoint to get the same rows as CSV.
Long runs
Run in the background, then poll.
A run that needs more time can go to the background: send "async": true, get 202 with the run id, and poll the runs endpoint until the status is done. Reading a run is free.
import os, time, requests
H = {"Authorization": f"Bearer {os.environ['QD_API_KEY']}"}
r = requests.post("https://api.quanticdata.io/v1/scraper/collectors/youtube_transcript/run", headers=H,
json={**{"videos":["dQw4w9WgXcQ"],"max_results":1}, "async": True}, timeout=30)
run_id = r.json()["payload"]["run_id"]
while True:
run = requests.get(f"https://api.quanticdata.io/v1/scraper/collectors/runs/{run_id}", headers=H, timeout=30).json()["payload"]
if run["status"] in ("done", "failed"):
break
time.sleep(5)
print(run["status"], run["count"])Errors
What can come back, and what is billed.
| Status | Meaning | Billed |
|---|---|---|
400 | The input does not match the collector (payload.errors) | No |
401 | Missing or wrong API key | No |
402 | Balance and free credit do not cover the run: top up | No |
422 | The input can never resolve; retrying the same input cannot help | No |
424 | The run failed at the source (blocked, timed out, layout changed); safe to retry | No |
429 | Rate limit or a per-account brake; wait for Retry-After | No |
Only delivered rows are billed: a run that delivers nothing costs nothing.
Price
$0.002 per delivered video.
| Rows | Cost |
|---|---|
| 1 | $0.002 |
| 1,000 | $2.00 |
| 10,000 | $20.00 |
Up to 50 videos per run; split larger jobs into several runs. The $2 free credit every month covers about 1,000 videos. No subscription; paid tiers lower the unit price. All collector prices: collector pricing.
Before you run it
Questions developers ask
Is there a free YouTube transcripts API?
Every account gets $2 of free usage each month, about 1,000 videos of this collector at $0.002 each, with no card.
Do I have to handle proxies, retries or blocks?
No. The collector runs on our network and handles them; a run blocked at the source returns 424 and is not billed.
How many rows can one run return?
Up to 50 videos per run (the max_results input, where the collector has one).
Can an AI agent run it?
Yes: the MCP server exposes it as the run_collector tool, with the same input and price. See the MCP server.
Run it on the free credit.
Create an API key, export it as QD_API_KEY and run the example. $0.002 per delivered video, no subscription.