npm package

YouTube transcripts in Vercel AI SDK

Add YouTube to any AI SDK model with three tools: transcript (optionally timestamped), YouTube search and a channel's video list. They run fine on Vercel and other serverless platforms, where direct YouTube requests are often blocked, and Mastra agents accept them as-is.

Get your free credits Setup steps

Set it up

1
Install
Alongside the AI SDK:
```
npm install @tubeagentkit/ai-sdk ai zod
```
2
Get an API key
Sign in at getyoutubetranscript.com and create a key on the developers page (100 free credits, no card). Keys start with sk_live_. Add it to your environment:
```
GETYOUTUBETRANSCRIPT_API_KEY=sk_live_...
```

Summarize a video with tools

import { generateText, isStepCount } from 'ai';
import { youtubeTranscript } from '@tubeagentkit/ai-sdk';

const { text } = await generateText({
  model: 'openai/gpt-5-mini',
  prompt: 'Summarize https://youtu.be/jNQXAC9IVRw with timestamps for each key point.',
  tools: {
    youtubeTranscript: youtubeTranscript(),
  },
  stopWhen: isStepCount(3),
});

console.log(text);

What you can build

A chat route that answers questions about any pasted YouTube link.
All tools at once with getYouTubeTranscriptTools(): transcript, search and channel videos.
Mastra: new Agent({ ..., tools: getYouTubeTranscriptTools() }) works without changes.

Good to know

The key is read when a tool first runs, so building the tools never throws. API errors are passed back to the model as tool errors.

Questions about Vercel AI SDK

Does it work with streamText and any provider?

Yes. These are standard AI SDK tools, so any model and any of generateText, streamText or agents can use them.

Does it work in Mastra?

Yes. Mastra 1.x agents accept AI SDK tools directly, so pass getYouTubeTranscriptTools() as the agent's tools.

What does it cost?

Each transcript, search or channel page uses one credit from your account, and failed requests are not charged. New accounts get 100 free credits, no card, and paid plans add more credits each month.