DEV Community: pantdeepakk5

How to get YouTube transcripts in n8n (and use them in an AI Agent)

pantdeepakk5 — Mon, 28 Sep 2026 14:22:29 +0000

If you want an n8n workflow to read what's actually said in a YouTube video (not just the title and description), you need a transcript. Here are two ways to get one: a built-in HTTP Request node hitting a REST API, or a community node that also plugs straight into an AI Agent as a tool.

Both need a free API key from GetYouTubeTranscript (100 credits, no card required). No Google API key or YouTube Data API quota involved.

Approach 1: HTTP Request node

If you don't want to install anything, n8n's built-in HTTP Request node can call the API directly, since it's a plain REST endpoint.

Get a key from the dashboard, then set it as an environment variable available to your n8n instance (don't hardcode it into the workflow JSON):

GETYOUTUBETRANSCRIPT_API_KEY=sk_live_...

Configure the HTTP Request node like this:

GET https://getyoutubetranscript.com/api/v1/transcript?v={{ $json.videoId }}
Authorization: Bearer {{ $env.GETYOUTUBETRANSCRIPT_API_KEY }}

Feed a video ID into the node's input and it returns the transcript, title, author, and thumbnail as JSON at $json.data.transcript.

There's a ready-to-import workflow for this on the GetYouTubeTranscript n8n guide if you'd rather not wire the node up by hand.

Processing more than a handful of videos

Feeding a list of video IDs straight into the HTTP Request node fires them all as fast as n8n can, which is fine for 5 videos and risky for 500 against a 60 req/min free-tier limit. Two built-in ways to pace it:

HTTP Request node → Add Option → Batching. Set Items per Batch and Batch Interval (ms). A batch size of 5 with a 5,000ms interval keeps you comfortably under a 60 req/min budget.
Split In Batches + Wait node. Strictly sequential, no parallel burst at all. Safer if you're close to the limit or a downstream API (an LLM summarizer, say) has its own throughput ceiling.

Also enable Continue On Fail on the HTTP Request node when processing a batch, since some videos will fail (private, no captions, a transient 429). Branch on the response with an IF node downstream instead of letting one bad video ID kill the whole run.

Approach 2: the community node

For a friendlier node UI (dropdowns instead of hand-built URLs) and AI Agent tool support, install n8n-nodes-getyoutubetranscript:

Settings → Community Nodes
Install a community node
Enter n8n-nodes-getyoutubetranscript and confirm

(Self-hosted only — community nodes aren't available on n8n Cloud. If you're on self-hosted via npm, npm install n8n-nodes-getyoutubetranscript and restart the instance instead.)

The node adds one credential type (a single API key field, tested against the free /resolve endpoint on save, so testing it never spends a credit) and eight operations:

Operation	Endpoint	Cost
Get Transcript	`GET /transcript`	1 credit
Search YouTube	`GET /search`	1 credit
Resolve Channel	`GET /resolve`	Free
Get Channel Latest Videos	`GET /channel/latest`	Free
Search Channel Videos	`GET /channel/search`	1 credit
List Channel Videos	`GET /channel/videos`	1 credit
List Playlist Videos	`GET /playlist`	1 credit
Get Credits	`GET /credits`	Free

Failed and rate-limited requests are never charged.

Using it as an AI Agent tool

Because it's a proper n8n node (not just an HTTP call), you can attach it directly to an AI Agent node as a tool. The agent decides when to call it, fills in the video URL itself, and reasons over whatever comes back:

That's useful for anything conversational: "summarize this video," "what does this channel's latest upload say about X," without you having to pre-wire which operation runs when.

Example workflow

The repo ships a ready-to-import workflow, summarize-youtube-video-with-ai.json: a Manual Trigger sets the video URL, the GetYouTubeTranscript node fetches the transcript, an HTTP Request node sends it to OpenAI's Chat Completions API to summarize, and a Set node formats the output. Import it via n8n's Import from File, then attach your GetYouTubeTranscript and OpenAI credentials.

Which one to use

HTTP Request node if you're on n8n Cloud (community nodes aren't supported there) or just need a quick one-off pull. The community node if you're self-hosted and want the dropdown UI, the extra operations (search, channels, playlists), or AI Agent tool support without hand-rolling the HTTP calls yourself.

Full API reference, error codes, and rate limits: getyoutubetranscript.com/docs.

Disclosure: I build GetYouTubeTranscript.

Get YouTube Transcripts Programmatically: A Free, Step-by-Step Developer Guide

pantdeepakk5 — Sun, 13 Sep 2026 08:30:26 +0000

There's no official YouTube transcript API. If you've ever needed a video's spoken text - for a RAG pipeline, a content research tool, an AI agent, or just to skim a 40-minute talk in 30 seconds - you've probably hit that wall.

Here's the free, step-by-step way to do it, with real requests you can run right now.

1. Get a free API key (no card required)

Sign up at getyoutubetranscript.com (Google sign-in) and grab a key from your dashboard, or provision one straight from a script/agent with no browser at all:

# Step 1: request a one-time code
curl -X POST https://www.getyoutubetranscript.com/api/v1/signup \
  -H "Content-Type: application/json" \
  -d '{"email": "you@example.com"}'

# Step 2: verify it, get your key back
curl -X POST https://www.getyoutubetranscript.com/api/v1/signup/verify \
  -H "Content-Type: application/json" \
  -d '{"email": "you@example.com", "otp": "123456"}'
# => { "success": true, "api_key": "sk_live_..." }

Either way you land with 100 free credits, 1 credit = 1 request, no expiry countdown, no card on file.

2. Fetch a transcript

curl "https://www.getyoutubetranscript.com/api/v1/transcript?v=dQw4w9WgXcQ" \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxx"

{
  "success": true,
  "data": {
    "video_id": "dQw4w9WgXcQ",
    "language_code": "en",
    "title": "...",
    "author_name": "...",
    "transcript": "We're no strangers to love...",
    "word_count": 1842
  }
}

v takes a raw video ID or a full URL - whichever's easier to pass around. Same call in Node:

const res = await fetch(
  'https://www.getyoutubetranscript.com/api/v1/transcript?v=dQw4w9WgXcQ',
  { headers: { Authorization: `Bearer ${process.env.YT_API_KEY}` } }
);
const { data } = await res.json();
console.log(data.transcript);

and Python:

import requests

res = requests.get(
    "https://www.getyoutubetranscript.com/api/v1/transcript",
    params={"v": "dQw4w9WgXcQ"},
    headers={"Authorization": f"Bearer {API_KEY}"},
)
print(res.json()["data"]["transcript"])

3. Search YouTube itself

Same key, a different endpoint - useful when you don't have a video ID yet:

curl "https://www.getyoutubetranscript.com/api/v1/search?q=rust+async&type=video&limit=10" \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxx"

Returns video_results (title, link, videoId, channel, views, published_date, description) or, with type=channel, matching channels instead. Pass the response's continuation_token back as page_token to page through more results.

4. Pull a whole channel

Two ways depending on whether a human or a script is driving:

No code, no signup: /youtube-channel-transcripts - paste a channel URL/@handle, pick up to 10 videos, download every transcript as one .txt. Free, capped at 10 per batch since it's proxying real scraping cost with no account behind it.

Scripted / more than 10: the same metered API, uncapped.

# List a channel's uploaded videos, paginated
curl "https://www.getyoutubetranscript.com/api/v1/channel/videos?channel=@channelname" \
  -H "Authorization: Bearer sk_live_xxxxxxxxxxxx"

{
  "success": true,
  "data": {
    "videos": [
      { "id": "abc123", "title": "...", "length": "10:37", "published_time": "3 days ago" }
    ],
    "has_more": true,
    "continuation_token": "opaque-token-pass-this-back-for-the-next-page"
  }
}

Loop video_ids through /api/v1/transcript and you've got a full-channel export. There's also /api/v1/channel/search (search inside one channel) and a free /api/v1/channel/latest for the home-tab upload shelf.

5. Whole playlists

/api/v1/playlist?list=<playlist_id> works the same way - paginated, same continuation_token convention, one call per page.

6. Bonus: skip the code entirely, give it to an AI agent

All of the above is also exposed as an MCP server (https://www.getyoutubetranscript.com/api/mcp), so Claude, Cursor, or any MCP-compatible client can call these tools directly instead of you writing the fetch calls. It supports both a plain API key and full OAuth 2.1 (dynamic client registration), so a client can connect without you ever pasting a secret into it. Example prompt once connected: "search this channel's videos for anything about pricing and summarize what they said."

Rate limits & what's next

Free tier: 60 requests/min, 100 credits total. Once you outgrow that, it's $5/month for 1,000 credits (or $4.50/mo billed annually), 200-300 req/min depending on plan - no feature gating, the full API is available from the free tier on.

Docs: getyoutubetranscript.com/docs
MCP + skill repos: github.com/tubeagentkit

If you build something with this, I'd genuinely like to see it - drop a comment.