YouTube Transcript API

Why I Built This

I was building a YouTube video summarizer and needed transcripts. The existing JavaScript libraries either required a YouTube API key or depended on Selenium for browser automation. I didn't want either - an API key adds friction for users, and Selenium is heavy for what should be a simple HTTP request.

So I looked at how YouTube actually serves transcripts (it's an undocumented XML endpoint), reverse-engineered the request flow, and built a library that fetches transcripts with plain HTTP calls. No API keys, no browser automation, no dependencies beyond Axios.

I also noticed the gap on npm - there wasn't a good JS/TS library for this. So I published it as a standalone package.

Tech Stack

• TypeScript - Type-safe development
• Axios - HTTP requests
• Commander.js - CLI framework
• http-proxy-agent - Proxy support
• Jest - Testing

Key Features

• No API key required - Works by fetching YouTube's undocumented transcript endpoint directly.
• Manual and auto-generated captions - Supports both caption types with language priority ordering.
• Four output formats - JSON (structured with timestamps), plain text, SRT, and WebVTT.
• Translation support - Translate transcripts to different languages when YouTube provides translations.
• CLI tool - Extract transcripts from the command line without writing code.
• Proxy support - Generic proxy and Webshare rotating proxy integration for high-volume use.
• Error handling - Specific errors for unavailable videos, disabled transcripts, and rate limiting.

What Came Next

This library became the transcript layer for Rundownly, a desktop YouTube summarizer I built with Tauri and Rust. The library handles transcript extraction; Rundownly adds AI summarization, timestamps, and a native desktop UI on top of it. What started as an npm package turned into the foundation for a full product.