Connector · The YouTube MCP
Let ChatGPT and Claude watch YouTube for you
Connect Whizbrew once and your AI reads YouTube live: any channel, any search, every video’s numbers, its transcript and its comments. You ask, it watches the videos so you don’t have to.
Connect your AI · 1-minute setup · $2 free
Is there a YouTube MCP for Claude and ChatGPT?
Yes. Whizbrew is a YouTube MCP: a connector you add to Claude, ChatGPT, Gemini or Cursor in a minute, no code and no API key. Your AI can then read public YouTube live: search results, channels and their videos, Shorts, each video’s views and likes, full transcripts with timestamps, comments, community posts and playlists.
Key takeaways
- The month’s top home coffee video, Speeed’s “I Tried Every Way to Make Coffee”, sets its challenge by 0:16. 2.4M views in 4 days.
- A 181K-subscriber channel beat its size 1.7×: tatsuki says “I finally got an espresso machine” within 10 seconds.
- The slowest open, 27 seconds on a book, did worst for its size: 188K views on 1.46M subscribers.
What your AI can read
- Search. Videos, Shorts, channels or playlists for any keyword, by relevance, date or views, uploaded in the last hour, week, month or year. “What are the most viewed air fryer videos uploaded this week?”
- Any channel. Subscribers, total views and video count, plus its About panel: country, join date, links and the business email it lists. “Which of these ten cooking channels list a business email?”
- A channel’s videos and Shorts. Newest or most viewed, with views, length and dates. Yours or a competitor’s. “What are @veritasium’s ten most viewed videos, and what do the titles have in common?”
- Any video, in detail. Title, description, views, likes, comment count, upload date, length, category and tags. “What tags and description does MrBeast use on his latest video?”
- Transcripts. The full captions of a video with timestamps, in the languages it has. Quote the hook, summarize a two-hour podcast, find the moment. “Summarize this podcast episode and give me the three best quotes with timestamps.”
- Comments. Top or newest comments with likes and replies, under videos and community posts. “What do people complain about under the latest iPhone reviews?”
- Community posts and playlists. A channel’s recent community posts with likes, its playlists, and the videos in any public playlist. “What did @LinusTechTips post in its community tab this month?”
- Suggestions and the Shorts feed. What YouTube suggests as people type, and the Shorts it pushes to a signed-out visitor in the US right now. “What does YouTube suggest when people type “how to start a”?”
What it doesn’t do
- It reads public YouTube data only. It doesn’t upload, edit, comment, like or subscribe for you.
- It doesn’t log into your channel, so it can’t see your private YouTube Studio analytics: watch time, retention, click-through rate, revenue. Your public videos, views and comments, yes, like anyone’s.
- Transcripts come from a video’s captions, the creator’s or YouTube’s automatic ones. A video without captions has no transcript.
- There’s no trending chart: your AI finds what’s hot by searching most viewed over a week or a month. Some counts, like comment likes, come rounded by YouTube.
The question to ask your AI
I run a small home coffee channel on YouTube. What are the top home espresso videos this month, and what do their first 30 seconds say? Show me views against each channel’s size, quote how each video opens, and read the top comments on the biggest one. Then tell me how to open my next video.
What your AI does, step by step
- Finds this month’s top videos. Searches YouTube for “home espresso”, most viewed first, only videos uploaded in the last month.
- Reads each video’s numbers. Opens the top six: exact views, likes, comments, upload date and the channel’s subscribers, so a small channel’s hit stands out.
- Reads the first 30 seconds. Pulls each video’s transcript and quotes 0:00 to 0:30, with timestamps, plus where the sponsor read lands.
- Reads the comments. The top comments under the most viewed video: what people laugh at, argue about and ask for next.
- Tells you how to open. What the winners say in their first sentence, what to keep out of the first 30 seconds, and what to film next.
A real run: Home espresso · YouTube, last 30 days, November 18, 2026
The first 30 seconds of the top six
Most viewed “home espresso” videos uploaded in the last month, quoted from their transcripts.
| Video | Views | First words | By 0:30 |
|---|---|---|---|
| I Tried Every Way to Make Coffee · Speeed | 2.44M | “There are a million people claiming that one of those ways makes the best cup of coffee.” | The challenge: a new method every morning for two weeks (0:16) |
| The Best Milk Foamer For Coffee · James Hoffmann | 978K | “Today, we’re going to take a look at the choices that you have to foam milk… at home.” | Who it’s for, and the three groups he’ll test (0:22) |
| Coffees at Home Have Lost the Plot · Edvasian | 665K | A stranger’s clip at 0:04: “Give my boyfriend his favorite bedtime drink.” | The verdict: “not a single drop of coffee in there” (0:20) |
| i finally got my dream espresso machine · tatsuki | 307K | “I finally got an espresso machine for home… pulling my first shot.” | The promise, then music and the unboxing |
| Breville Finally Updated Their Most Popular Espresso Machine · Lance Hedrick | 236K | “Breville’s most popular espresso machine has finally got an upgrade.” | The news, then why Breville waited |
| This VIRAL Espresso Machine Is Very Confusing · Morgan Eckroth | 188K | “At long last, my first book, Coffee for Here, is officially out.” | Still on the book. The machine arrives at 0:50 |
The month’s most viewed coffee-at-home videos
YouTube, uploaded 18 Oct – 18 Nov, exact counts.
- @SpeeedCo: “I Tried Every Way to Make Coffee” · 2,436,670 views · November 14, 2026
- @jameshoffmann: “The Best Milk Foamer For Coffee” · 977,645 views · October 28, 2026
- @Edvasian: “Coffees at Home Have Lost the Plot” · 664,619 views · November 11, 2026
- @havenotats: “i finally got my dream espresso machine” · 307,452 views · October 31, 2026
Views against channel size
Each video’s views ÷ its channel’s subscribers.
- tatsuki, dream machine: 1.7× (307K views from 181K subscribers)
- Speeed, every method: 0.92× (2.44M views from 2.66M subscribers)
- Lance Hedrick, Breville: 0.51× (236K views from 462K subscribers)
- Morgan Eckroth, viral machine: 0.13× (188K views from 1.46M subscribers)
The takeaway: Five of the six say what the video is about in their first five seconds. The one that opened on a book plug did the worst for its size by far.
What viewers say under the #1 video
Top comments on “I Tried Every Way to Make Coffee”, about 3,300 in all.
| Comment | Likes |
|---|---|
| “We all knew the coffee episode was inevitable” | 2.6K |
| “Turkish coffee aura definitely goes up if you’re cooking it in hot sand.” | 2.3K |
| “…with coffee we’ve finally completed the modern day hipster starter pack” | 2.1K |
| “Me, a totally aura-less middle aged lady… I’ll always use my Aeropress.” | 1.4K |
| “JUSTICE FOR THE MOKA POT” | 1K |
| “Id actually love a part 2 collab with a coffee expert…” | 901 |
Where the sponsor read lands
From the transcripts and the video chapters.
- Sponsored: 4 of 6 (Cometeer ×2, Standart, LELO)
- In the first 30 seconds: 0 (None of the six)
- Earliest read: 1:56 (Edvasian, after the hook)
- Speeed’s read: 30:41 (Three quarters into 42 minutes)
The takeaway: The top comments pick sides: AeroPress, moka pot, a part 2 with an expert. A comparison people can argue with brings the comments, and the next video idea with them.
The plan your AI writes
- Say the video in your first sentence. Lance opens on the news (“Breville’s most popular espresso machine has finally got an upgrade”), Hoffmann on who it’s for. Speeed sets its two-week challenge by 0:16 and got 2.44M views in 4 days.
- Or open on someone else’s clip. Edvasian starts on a stranger’s “bedtime drink” at 0:04 and lands the joke by 0:20: 665K views in a week. Reacting is a format, and coffee TikToks are its raw material.
- Keep the plugs past the hook. None of the six sells in its first 30 seconds; the earliest sponsor read is at 1:56. Morgan Eckroth opened on her book for 27 seconds: 188K views on 1.46M subscribers, the weakest ratio of the six.
- Film the comparison people argue about. “Every way to make coffee” drew 3,300 comments, and its top ones defend the AeroPress and the moka pot. Make that part 2 (901 people liked the idea), or film your first week with a machine: tatsuki did 1.7× his size with it.
How this was made. Asked once in Claude with Whizbrew connected, on 18 November 2026. It searched YouTube for “home espresso”, most viewed, uploaded in the last month (20 results), and kept the six top creator videos about coffee at home (set aside: a video about appliances in general and three brand videos). It read each one’s exact views, likes, comments, date and channel subscribers, pulled the six transcripts and quoted 0:00 to 0:30, and read the top 20 comments under the most viewed one. Cost of the whole run: about 4 cents.
Without live data
I can’t watch YouTube or read its transcripts, so I can’t tell you which espresso videos did well this month or how they open. In general, a strong YouTube intro hooks viewers in the first few seconds: ask a question, tease the result, keep it short and add a clear call to action…
Questions
What is a YouTube MCP?
MCP (Model Context Protocol) is the standard way to plug a data source into an AI. A YouTube MCP lets ChatGPT, Claude or another AI read YouTube while it answers you: real videos, transcripts, comments and numbers instead of guesses. Unlike the YouTube API, there’s no key, no quota and no code. Whizbrew is one, with TikTok, Instagram, ad libraries and more in the same connection.
How do I connect YouTube to Claude?
Add Whizbrew as a connector in Claude’s settings and sign in with your Whizbrew account. It takes about a minute, no code and no API key; the exact clicks are in the setup section above. It works the same in Claude Code. Then ask about YouTube in plain words.
Does it work in ChatGPT?
Yes. The same connector works in ChatGPT, Claude, Gemini, Cursor, Claude Code and Codex, with the same YouTube data. Pick your AI in the setup section for its steps.
Can it read YouTube transcripts?
Yes. Your AI can read the full captions of any public video that has them, with timestamps, and quote, summarize or search them. This page’s run read six transcripts to compare how the top coffee videos open. A video with no captions has no transcript to read.
Can it read my YouTube Studio analytics or upload videos?
No. It only reads public data, so it doesn’t upload, edit or reply, and it never logs into your channel: no watch time, retention or revenue from YouTube Studio. It can read your public channel, videos and comments, the same way it reads anyone’s.
Is it free? What does it cost?
You get $2 of free credit when you sign up. Each lookup costs a fraction of a cent, and lookups that find nothing are free. The whole run on this page, six transcripts included, cost about 4 cents.
Are you an AI agent? Start here.
You can open a Whizbrew account on your own: no email, no sign-up form, no key to paste. One request gives you a key that works at once, with free credit. Your human claims the account later.
# 1. Read what Whizbrew is, what a call costs and how to connect
curl -s https://whizbrew.com/llms.txt
# 2. Open your own account: one request, no email
curl -s -X POST https://whizbrew.com/v1/agent/start \
-H 'content-type: application/json' \
-d '{"agent":"<your name>"}'
# 3. Your human claims the account within 7 days. The whole flow:
curl -s https://whizbrew.com/agents.md
- Start without an account: https://whizbrew.com/agents.md
- Every tool: https://whizbrew.com/tools.md
- API reference: https://whizbrew.com/openapi.json
- MCP server: https://whizbrew.com/mcp
Not an agent? Paste this line into yours: Set up Whizbrew for me: read https://whizbrew.com/llms.txt and follow it.