langchain-getyoutubetranscript

YouTube transcript tools and a document loader for LangChain, powered by the GetYouTubeTranscript API. Give agents the transcript of any YouTube video (optionally with [m:ss] timestamps) and YouTube search, or load transcripts into RAG pipelines.

The API fetches transcripts on its own servers, so it works from cloud servers and serverless functions without proxies, and without RequestBlocked / IpBlocked errors.

Install

pip install langchain-getyoutubetranscript

Get an API key at getyoutubetranscript.com/dashboard (free tier included) and set it:

export GETYOUTUBETRANSCRIPT_API_KEY=sk_live_...

Or pass api_key="..." to any class.

Tools

from langchain_getyoutubetranscript import GetYouTubeTranscriptTool

tool = GetYouTubeTranscriptTool()
print(tool.invoke({"video": "https://youtu.be/jNQXAC9IVRw", "timestamps": True}))
# Title: Me at the zoo
# Channel: jawed
# Language: en
#
# [0:01] All right, so here we are, in front of the elephants
# ...

API errors (no captions, invalid video, out of credits) come back to the agent as a short message instead of raising.

With an agent

from langchain.agents import create_agent
from langchain_getyoutubetranscript import GetYouTubeTranscriptSearchTool, GetYouTubeTranscriptTool

agent = create_agent(
    model="anthropic:claude-sonnet-4-5",
    tools=[GetYouTubeTranscriptSearchTool(), GetYouTubeTranscriptTool()],
)
agent.invoke(
    {"messages": [{"role": "user", "content": "Find a short talk on transformers and summarize it with timestamps."}]}
)

Document loader

from langchain_getyoutubetranscript import GetYouTubeTranscriptLoader

docs = GetYouTubeTranscriptLoader(
    ["https://youtu.be/jNQXAC9IVRw", "5e37ZT3SQbk"],
    language="en",
    timestamps=False,
).load()

print(docs[0].metadata)
# {'source': 'https://www.youtube.com/watch?v=jNQXAC9IVRw', 'video_id': 'jNQXAC9IVRw',
#  'title': 'Me at the zoo', 'author_name': 'jawed', 'language_code': 'en', 'word_count': 39}

One Document per video. With timestamps=True, page_content is [m:ss] lines and metadata["segments"] holds {start, duration, text} per caption line (seconds).

Pricing

Each transcript or search request uses one credit from your GetYouTubeTranscript account. Failed requests are not charged.

Development

pip install -e ".[test]"
pytest                                        # unit + LangChain standard tests, no network
GETYOUTUBETRANSCRIPT_API_KEY=... pytest tests/integration_tests   # live, spends credits

Links

• GetYouTubeTranscript API docs
• Python SDK (this package is built on it)
• License: MIT