How to get YouTube transcripts in Python

import os
import requests

BASE = "https://lovelyrouter.com/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['LOVELY_API_KEY']}"}


def run_tool(tool, tool_input, max_results):
    res = requests.post(
        f"{BASE}/tools/{tool}/run",
        headers=HEADERS,
        json={"input": tool_input, "max_results": max_results},
        timeout=90,
    )
    run = res.json()
    if not res.ok:
        raise RuntimeError(run["error"]["message"])

    # 202 = still running: poll, letting the server wait up to 60 s per request.
    while res.status_code == 202:
        res = requests.get(f"{BASE}/tools/runs/{run['id']}", headers=HEADERS, params={"wait": 60}, timeout=90)
        res.raise_for_status()
        run = res.json()

    print(f"{tool}: {run['status']}, {run['results']} results, charged {run['charge_usd']} USD")
    if run["error"]:
        print("  error:", run["error"])
    return run["items"] or []


items = run_tool(
    "youtube-transcript",
    {"startUrls": ["https://www.youtube.com/watch?v=dQw4w9WgXcQ"], "lang": "en"},
    max_results=1,
)
video = items[0]
print(video["selected"]["title"])
for seg in video["transcript"][:5]:
    print(seg["startTime"], seg["text"])

# Up to 30 videos from a channel (a handle, or a channel URL in startUrls).
videos = run_tool("youtube-channel", {"youtubeHandles": ["@MrBeast"]}, max_results=30)

# Or long videos about a topic uploaded this year.
# videos = run_tool("youtube-search", {"keywords": ["espresso tutorial"], "duration": "l", "uploadDate": "y"}, max_results=30)

urls = [v["url"] for v in videos if "/watch?v=" in v.get("url", "")]


def transcripts(urls, lang="en"):
    out = []
    for i in range(0, len(urls), 50):  # the tool takes up to 50 links per run
        batch = urls[i : i + 50]
        out += run_tool("youtube-transcript", {"startUrls": batch, "lang": lang}, max_results=len(batch))
    return out


found = transcripts(urls)
missing = set(urls) - {t["inputSource"] for t in found if t.get("transcript")}
print(f"{len(urls) - len(missing)} transcripts, {len(missing)} videos without one")

import anthropic

llm = anthropic.Anthropic(base_url="https://lovelyrouter.com", api_key=os.environ["LOVELY_API_KEY"])


def summarize(item):
    text = "\n".join(f"[{s['startTime']}] {s['text']}" for s in item["transcript"])
    message = llm.messages.create(
        model="claude-haiku-5.5",
        max_tokens=1024,
        messages=[{
            "role": "user",
            "content": "Summarize this video transcript in five bullet points. "
                       "Start each bullet with the timestamp where the point is made.\n\n" + text,
        }],
    )
    return message.content[0].text


for item in found:
    if item.get("transcript"):
        print(item["inputSource"])
        print(summarize(item), "\n")