6 Best YouTube MCP Servers, Compared

Looking for the official YouTube MCP server? There isn't one. Google doesn't publish one and neither do the authors of the protocol, so every option you'll find is a community project or a vendor's hosted service.

That sounds worse than it is. The servers split cleanly into three groups, and once you know which group you need, the choice takes about a minute. A YouTube MCP server is just the connector that plugs YouTube into an AI app through the Model Context Protocol — and because they all speak the same standard, swapping one for another costs you nothing.

What are the main approaches to YouTube MCP servers?

Three, and they differ by how much access they need rather than by quality.

Transcript-first. Fetches captions so your assistant can summarise and analyse. Usually no API key at all.

API-based data. Search, metadata, statistics, trending. Needs your own YouTube Data API key from Google Cloud.

Write-capable. Uploads, playlists, comments, captions. Needs OAuth rather than a key, because it acts as you.

There's a smaller fourth group too: hosted servers that read your own channel's private analytics. More on those below.

How does a transcript-first server work?

It reads captions the way your browser does — and there's a good reason it doesn't use the official API.

Google's captions.download endpoint requires OAuth with the youtube.force-ssl scope, and in practice only a video's owner can download its caption track. An API key alone returns forbidden (403), every time. So if a server wants to fetch captions from videos you don't own, scraping is the only route left.

Is that fragile? Yes, and the projects say so themselves. The widely used youtube-transcript-api library warns that it relies on an undocumented part of the API called by the web client, with no guarantee it keeps working tomorrow. Worth knowing before you build a workflow on it.

When do you need an API-based server?

The moment you want search, statistics or trending data rather than just words.

That means your own key, and it means living inside Google's quota. Each project gets 10,000 units a day for most endpoints, plus separate buckets of 100 search.list calls and 100 videos.insert calls. Each of those costs one unit from its own bucket, so your real ceiling is a hundred searches a day.

One number catches people out: captions.list costs 50 units against the 10,000 pool — fifty times a videos.list. If a caption-heavy workflow drains your quota by lunchtime, that's usually why. Quotas reset at midnight Pacific Time, and even invalid requests cost at least a point.

Which YouTube MCP servers are worth using?

These six, all MIT-licensed open source, ordered by how much access they need.

• kimtaeyoon83/mcp-server-youtube-transcript (TypeScript) — one job, done simply: get_transcript from a normal URL, a Shorts URL or a bare video ID. No API key.
• jkawamoto/mcp-youtube-transcript (Python) — transcripts plus get_timed_transcript and get_available_languages. Start here if you work across languages. No API key.
• anaisbetts/mcp-youtube (TypeScript) — pulls subtitles via yt-dlp. Handy when captions are awkward, at the cost of an external binary. No API key.
• ZubeidHendricks/youtube-mcp-server (TypeScript) — ten tools: video, channel and playlist retrieval, search and transcripts. Needs a YOUTUBE_API_KEY.
• anwerj/youtube-uploader-mcp (Go) — write access: uploads, video updates, thumbnails, captions, playlists. OAuth.
• pauling-ai/youtube-mcp-server (Python) — the biggest surface, around forty tools across the Data, Analytics and Reporting APIs. Write access, OAuth.

Disclosure: we make one too. VidTraffic's YouTube MCP server is a hosted, read-only endpoint that hands your agent your own channel's analytics — retention curves, traffic mix, search phrases, the questions under your videos. No API key to manage, and it never posts, edits or deletes. It solves a different problem from the six above: they read public videos, this reads your private performance data. Judge it on that, not as a transcript tool.

One caution before you adopt any repo. The official MCP registry lists dozens of YouTube entries, and listing is not endorsement. Check the LICENSE file rather than trusting a badge in the README — several popular repos claim MIT without actually including a licence.

How safe is it to connect one?

The useful question isn't "is this trustworthy?" but "what could it do if it weren't?"

Prefer read-only. A transcript server that can't write can't damage a channel. Worst case, it fetches something you didn't want.

Treat an API key like a password. Create it for this purpose, restrict it in Google Cloud, never paste it into a chat. Only a server running on your own machine should hold it.

Know what OAuth grants. A write-capable server acts as you — it can publish, edit and delete. Give that only to code you've read, and keep a human approving anything destructive.

Skim the source. Most of these are small projects by individuals. That's not a reason to avoid them, but it is a reason to look before handing over credentials.

Which YouTube MCP server should you choose?

Whichever one solves today's problem with the least access. Really — start small.

Want summaries? Take a transcript-first server and stop there. Want to research other channels' public numbers? Add an API-based server and a key. Only reach for OAuth and write access when you actually intend to let software act on your channel. And if what you really wanted was your own analytics, a hosted read-only server saves you the Google Cloud project entirely.

Which server suits which creator?

• The researcher, summarising and comparing videos: a transcript-first server, nothing more.
• The competitor-watcher, tracking other channels: an API-based server with your own key.
• The analyst, asking why a video underperformed: a hosted analytics server on your own data.
• The operator, automating uploads and playlists: a write-capable OAuth server, with approvals.

Most creators never need to leave the first or third line.

How do you handle transcripts in multiple languages?

By checking what exists before you ask for it.

Captions carry a BCP-47 language tag, and tracks made by speech recognition are marked ASR, so a server can tell you what's available rather than guessing. The common pattern — the one youtube-transcript-api uses — is to list the transcripts, then request languages in descending priority: ask for German, fall back to English. Manually written captions win over automatic ones by default.

Where a track is marked translatable, you can also ask YouTube to translate it rather than translating the text yourself. Servers that expose an "available languages" tool make all of this visible. Servers that only expose get_transcript leave you guessing.

Why do you get 403 Forbidden errors?

Almost always because of the official API, and there are three causes worth telling apart.

quotaExceeded — you've spent the day's units. Wait for the midnight Pacific reset, or work out which tool is burning quota. Caption and search calls are the usual suspects.

forbidden — the request wasn't properly authorised. For caption downloads this is expected rather than a bug: the endpoint needs OAuth and owner permissions, so an API key will never satisfy it.

accessNotConfigured — the YouTube Data API isn't enabled on your Google Cloud project. Enable it and retry.

And here's what is not a 403: cloud IP blocking. YouTube blocks addresses belonging to AWS, Google Cloud and Azure, so a scraping-based server that works on your laptop often fails once deployed — but that shows up as HTTP 429 or a bot-detection error. Most write-ups conflate the two. If you're seeing 429s, the fix is where your code runs, not your credentials.

Frequently asked questions

Is there an official YouTube MCP server?

No. The reference servers maintained by the protocol's authors don't include YouTube, and Google doesn't publish one. Everything available is third-party.

Do YouTube MCP servers need an API key?

Transcript-first servers usually don't. Anything doing search, statistics or trending needs your own YouTube Data API key, and anything that writes needs OAuth.

Are YouTube MCP servers free?

The open-source ones are free to run. Costs come from hosting, or from YouTube Data API usage beyond the free quota.

Can an MCP server manage my channel?

Some can — uploads, playlists, comments, captions — but only with OAuth. Grant write access deliberately, and keep a person approving anything destructive.

Sources

• Model Context Protocol — introduction
• MCP reference servers
• YouTube Data API — getting started and quota
• YouTube Data API — quota cost calculator
• YouTube Data API — error reference
• YouTube Data API — captions resource
• youtube-transcript-api documentation

Setting one up for the first time? The walkthrough for connecting YouTube to Claude covers it end to end.