Video transcripts in LangChain

Short answer: there are two ways. For indexing, a small loader turns a transcript into LangChain Document objects, one per caption segment with its start time. For agents, the Transcript Dock MCP server plugs into LangChain through langchain-mcp-adapters and the agent fetches transcripts itself.

A document loader#

Twenty lines, no extra package. Each Document keeps the video URL and timing, so retrieval results can cite the moment they came from. Group segments into larger chunks before embedding, as shown in transcripts for RAG.

loader.py

# pip install langchain-core requests
import requests
from langchain_core.documents import Document
 
API = "https://www.transcriptdock.com"
H = {"Authorization": "Bearer td_live_YOUR_KEY"}
 
def load_video(url: str, mode: str = "captions_only") -> list[Document]:
    """One Document per caption segment, with the video URL and start time as metadata."""
    job = requests.post(f"{API}/v1/jobs", headers={**H, "Idempotency-Key": f"lc-{url}"[:128]},
        json={"source": {"url": url}, "mode": mode}).json()
    while job["status"] not in ("succeeded", "failed", "cancelled"):
        job = requests.get(f"{API}/v1/jobs/{job['id']}", params={"wait": 25}, headers=H).json()
    if job["status"] != "succeeded":
        raise RuntimeError(job["error"]["code"])
    t = requests.get(f"{API}/v1/transcripts/{job['result_id']}", headers=H).json()
    meta = {"source": t["source"]["canonical_url"], "title": t["source"]["title"], "language": t["language"]}
    return [Document(page_content=s["text"], metadata={**meta, "start": s["start"], "end": s["end"]})
            for s in t["segments"]]
 
docs = load_video("https://www.youtube.com/watch?v=dQw4w9WgXcQ")

Agent tools over MCP#

MultiServerMCPClient connects to the remote server and turns its tools into LangChain tools: get_video_transcript, YouTube search, channel and playlist listing and TikTok profile listing.

agent.py

# pip install langchain-mcp-adapters langgraph langchain-anthropic
import asyncio
from langchain_mcp_adapters.client import MultiServerMCPClient
from langgraph.prebuilt import create_react_agent
 
async def main():
    client = MultiServerMCPClient({
        "transcriptdock": {
            "url": "https://www.transcriptdock.com/mcp",
            "transport": "streamable_http",
            "headers": {"Authorization": "Bearer td_live_YOUR_KEY"},
        }
    })
    tools = await client.get_tools()
    agent = create_react_agent("anthropic:claude-sonnet-5-5", tools)
    out = await agent.ainvoke({"messages": "Summarize https://www.youtube.com/watch?v=dQw4w9WgXcQ"})
    print(out["messages"][-1].content)
 
asyncio.run(main())

Tool names, arguments and credit costs are on the MCP page. LlamaIndex and other frameworks that speak MCP connect the same way.

More guides

How to get a YouTube transcript

Five ways to get the transcript of a YouTube video: in YouTube itself, with yt-dlp, in Python, through the Transcript Dock API, or by asking Claude.

How to get a TikTok transcript

Get the transcript of a TikTok video as text or subtitles, list a profile's latest videos, and handle videos without captions.

Video transcripts for RAG and LLM pipelines

Turn YouTube and TikTok videos into chunks an LLM can search and cite: fetch timed segments, chunk by time, keep timestamp links.

YouTube and TikTok transcripts in n8n

Fetch video transcripts in an n8n workflow with three HTTP Request nodes, or give an n8n AI agent the Transcript Dock MCP tools.