Deploy to Cloudflare
Serve transcripts from a Cloudflare Worker — the endpoint the React components talk to, with caching and the edge proxy caveat.
extract-youtube has no filesystem access and no native modules, so it runs on Cloudflare Workers as-is. This page builds the transcript endpoint the React components expect.
A transcript Worker
Create the Worker and install the package
npm create cloudflare@latest transcripts -- --type hello-world --lang ts
cd transcripts
npm install extract-youtubeEnable Node compatibility
The package's optional proxy support imports Node built-ins, so turn on nodejs_compat in wrangler.jsonc:
{
"name": "transcripts",
"main": "src/index.ts",
"compatibility_date": "2026-09-01",
"compatibility_flags": ["nodejs_compat"]
}Write the handler
import { YouTubeTranscriptApi, extractVideoId } from 'extract-youtube';
const api = new YouTubeTranscriptApi();
export default {
async fetch(request: Request): Promise<Response> {
const raw = new URL(request.url).searchParams.get('videoId') ?? '';
const videoId = extractVideoId(raw);
if (!videoId) return Response.json({ error: 'Missing videoId' }, { status: 400 });
try {
const transcript = await api.fetchTranscript(videoId, { languages: ['en'] });
return Response.json({ videoId, snippets: transcript.toRawData() });
} catch (error) {
// 200 + `error`: the React components treat this as "no captions", not a crash.
return Response.json({
videoId,
snippets: [],
error: error instanceof Error ? error.message : 'Failed to fetch transcript',
});
}
},
} satisfies ExportedHandler;Deploy
npx wrangler deployPoint the components at it: transcriptUrl="https://transcripts.<your-subdomain>.workers.dev". If the page and the Worker are on different origins, add an Access-Control-Allow-Origin header to the responses.
Cache transcripts
Each request to YouTube costs time, and a proxy costs money. Put a cache in front — the Workers Cache API needs no bindings:
export default {
async fetch(request: Request, _env: unknown, ctx: ExecutionContext): Promise<Response> {
const cache = caches.default;
const hit = await cache.match(request);
if (hit) return hit;
const response = await handle(request); // the handler above
const body = await response.clone().json<{ error?: string }>();
// Transcripts don't change: keep them a day. Keep failures briefly — a block is transient.
response.headers.set('Cache-Control', body.error ? 'public, max-age=300' : 'public, max-age=86400');
ctx.waitUntil(cache.put(request, response.clone()));
return response;
},
};For a cache that survives across data centres, store toRawData() in KV or R2 keyed by video ID instead.
Proxies on Workers
proxyConfig does nothing on Workers: it relies on a Node socket agent, which the Workers runtime ignores, so requests silently go out direct. When YouTube starts blocking the Worker's IPs, use a fetch-through proxy injected as a custom httpClient — see Proxies on edge runtimes.
Other serverless platforms
The same handler works anywhere fetch does:
import { YouTubeTranscriptApi } from 'extract-youtube';
export const handler = async (event: { videoId: string }) => {
const transcript = await new YouTubeTranscriptApi().fetch(event.videoId);
return { statusCode: 200, body: JSON.stringify(transcript.toRawData()) };
};import { YouTubeTranscriptApi } from 'extract-youtube';
export const config = { runtime: 'edge' };
export default async function handler(req: Request) {
const videoId = new URL(req.url).searchParams.get('videoId')!;
const transcript = await new YouTubeTranscriptApi().fetch(videoId);
return Response.json(transcript.toRawData());
}A video library on the same Worker
The library admin API from extract-youtube/library runs on Workers too, over D1. See Library admin API for the setup, and the live demo for a site that serves transcripts, the library API and the app together.
Proxies
Keep fetching when YouTube rate-limits your server — generic proxies, Webshare residential IPs, and fetch-through proxies on edge runtimes.
Floating player
A draggable, resizable YouTube player that keeps playing while people move around your app — with picture-in-picture, a queue, resume and synced captions.