How to set up a YouTube MCP server for Claude Code

You can now paste a YouTube link into Claude Code and get a full transcript back. No browser tabs, no copy-pasting, no fragile scraping scripts.

The trick is a YouTube MCP server powered by TranscriptAPI. It connects Claude Code to YouTube's transcript data through the Model Context Protocol. The whole setup takes less than 10 minutes. Here's exactly how to do it, plus what you can build once it's running.

What an MCP server actually does

MCP stands for Model Context Protocol. It's an open standard that lets AI tools like Claude Code talk to external services. GitHub, databases, browsers, YouTube — all fair game.

Think of it like a plugin system. You install a server, Claude Code detects it, and your AI assistant picks up a new capability it didn't have before.

Here's how the architecture works: Claude Code acts as the MCP client. The server wraps the access logic for whatever external resource you're connecting to. They communicate through one of three transport modes — stdio, SSE, or Streamable HTTP. You don't need to think about that part. The CLI handles it.

As of early 2026, the MCP ecosystem has over 200 servers available. YouTube transcript servers are among the most requested because video content is one of the hardest data sources to work with programmatically. And there's a lot of it.

For the bigger picture on why agents need YouTube access, read our guide to giving AI agents YouTube data via MCP.

Why YouTube transcripts matter for developers

YouTube hosts over 800 million videos. A huge chunk of technical knowledge lives exclusively in video format. Conference talks, tutorials, product demos, podcasts.

But video is slow to consume. A 45-minute conference talk might contain three useful insights buried in 42 minutes of filler. Transcripts let you search, skim, and extract what you need in seconds instead of minutes.

That number surprised me when I first looked into it. According to Domo's Data Never Sleeps report, YouTube users upload over 500 hours of video every minute. The amount of knowledge locked inside video that never gets indexed as text is staggering.

With a YouTube MCP server in Claude Code, you can:

• Summarize a 2-hour tutorial into bullet points

• Compare how three different creators explain the same concept

• Extract every code snippet mentioned in a video

• Pull quotes with timestamps for reference

• Search YouTube and analyze results without leaving your terminal

• Analyze an entire playlist or channel in one conversation

That last one is the real unlock. Teams are using this to process 20+ videos from a competitor's YouTube channel in a single Claude Code session. Try doing that manually.

Pick your setup method

TranscriptAPI gives you two ways to connect YouTube data to Claude Code. Each fits a different workflow.

Want to use Claude.ai or Claude Desktop instead? See our Claude Desktop MCP setup guide.

Option 1: MCP server (for Claude.ai and Claude Desktop)

TranscriptAPI runs a hosted MCP server at https://transcriptapi.com/mcp. You connect through OAuth, and Claude handles everything in the background. Zero infrastructure on your end.

Best for: teams using Claude's web or desktop interface who want automatic transcript fetching when they paste a YouTube URL.

Option 2: Agent skills (for Claude Code, Cursor, Windsurf, and other agents)

TranscriptAPI also publishes a set of YouTube skills that install directly into Claude Code and other AI coding agents. These give you 12 specialized tools covering transcripts, search, channels, playlists, and captions.

Best for: developers working inside Claude Code or other agent-based tools who want the full YouTube toolkit.

I'll walk through both. Pick whichever matches your setup.

Setting up the MCP server (Claude.ai / Claude Desktop)

You'll need an active TranscriptAPI account and a Claude.ai account on any tier.

Step 1: Add the connector

Open Claude's settings and navigate to the MCP connectors section. Click "Add custom connector" and enter these two fields:

• Name: TranscriptAPI

• Server URL: https://transcriptapi.com/mcp

Copy the server URL directly from your TranscriptAPI dashboard to avoid typos. One wrong character and the connection silently fails.

Step 2: Authorize the connection

Claude redirects you to TranscriptAPI's OAuth page. Sign in and confirm that you want Claude to access your account.

One thing to know: each transcript fetch uses API credits from your TranscriptAPI balance. Browsing a channel's video list is free, but pulling the actual transcript costs credits. Not a huge deal for occasional use, but worth tracking if you're doing batch analysis.

Step 3: Set permissions

Set the permission to "Allow unsupervised." This lets Claude fetch transcripts automatically when it detects a YouTube URL in your conversation.

Without this, you get a permission prompt every single time. Gets old fast when you're processing multiple videos.

Step 4: Verify it works

Open a new conversation, paste any YouTube link, and ask a question about the video. If the integration is working, Claude fetches the transcript behind the scenes and responds based on the content.

No special syntax needed. Just paste the URL and talk normally. That's the whole point of MCP.

Setting up YouTube skills (Claude Code / Cursor / Windsurf)

This approach installs TranscriptAPI's YouTube skills directly into your coding agent. You don't need a separate MCP configuration.

Step 1: Install the skills

Run one command in your terminal:

bash

npx skills add ZeroPointRepo/youtube-skills --skill youtube-full

This installs the youtube-full skill, which bundles all 12 YouTube tools into a single package. Transcripts, search, channel browsing, playlists, captions. Everything.

Step 2: Get your API key

On first use, the skill walks you through a free registration. It asks for your email, sends an OTP code, and saves the API key to your agent config automatically.

The free tier includes 100 credits at signup, which is enough for about 100 transcript fetches. Paid plans start at $5/month for 1,000 credits if you need more volume.

Step 3: Verify the connection

Start Claude Code and ask it to fetch a transcript from any YouTube video. If the skill is installed correctly, Claude will call the TranscriptAPI backend and return the full text.

You can also install individual skills if you only need specific features:

SkillWhat it doestranscriptVideo transcript extractionyoutube-searchSearch YouTube videos and channelsyoutube-channelsBrowse channel uploads and latest videosyoutube-playlistExtract full playlist contentscaptionsGet captions and subtitles in multiple languages

Install any of these individually with npx skills add ZeroPointRepo/youtube-skills --skill [name].

What you can do once it's running

TranscriptAPI processes over 6 million transcripts per month with 99.9% uptime. Here's what that infrastructure unlocks inside Claude Code.

Research and summarization

Paste a link and ask Claude to summarize the key points. This works especially well with educational content, conference talks, and podcast episodes where the signal-to-noise ratio varies.

Claude can pull out the main arguments, identify action items, and flag sections where the speaker contradicts earlier points. It's like having a research assistant who watches videos at infinite speed.

Ever spent 20 minutes scrubbing through a keynote looking for one specific quote? That's the kind of thing that disappears from your workflow once transcripts are searchable.

Competitive analysis

Point Claude at a competitor's YouTube channel and have it analyze their content strategy. What topics do they cover most? How often do they post? What questions keep coming up?

TranscriptAPI handles up to 10 videos simultaneously. One team I talked to analyzed 50 competitor videos in an afternoon using batch requests through the playlist skill. How long would that take with manual copy-paste? Days, probably.

Content repurposing

Got a 30-minute webinar you need to turn into a blog post? Feed the transcript to Claude with instructions about tone and format. The first draft won't be perfect, but it gets you 70% of the way there in about 30 seconds.

Make sense? This works for turning videos into newsletters, social posts, documentation, or training materials too.

YouTube search without leaving your editor

The youtube-search skill lets Claude search YouTube directly. Ask it to find videos on a topic, and it returns results you can then pull transcripts from. No browser needed. No context switching.

This is the workflow that saves the most time for research-heavy projects.

Code extraction from tutorials

Technical tutorials on YouTube are packed with code, but it's trapped in video format. Claude can pull every snippet from the transcript, organize them in execution order, and flag dependencies or setup steps the presenter mentioned.

Way faster than pausing and typing out code from a screen recording. Sound familiar?

Handling long videos and edge cases

So you've got the setup working. What happens when you throw something tricky at it?

Token limits on long content

A 2-hour video transcript can be massive. Two workarounds:

1. Request the transcript without timestamps (saves 20-30% of tokens immediately)

2. Ask Claude to analyze specific sections instead of pulling the full transcript

TranscriptAPI handles long videos well, though transcripts over 2 hours may take 10-30 seconds to process.

Videos without captions

Not every YouTube video has transcripts available. Auto-generated captions cover most English content, but some creators disable captions or post in languages without auto-caption support. When this happens, the tool returns an error and Claude will let you know.

Use the captions or get-transcript-languages tool first to check availability before making a request.

Private and restricted videos

Age-restricted, private, and regionally blocked videos won't work. The server can only access publicly available content. No workaround for this one.

Why not just scrape YouTube directly?

You might wonder: why use an API instead of scraping transcripts yourself?

The honest answer: YouTube actively blocks transcript requests from major cloud IP ranges. Tools that rely on yt-dlp or headless browsers break regularly whenever YouTube changes their frontend. TranscriptAPI handles all the infrastructure, rate limiting, and IP rotation behind the scenes. You just make the request and get text back. No maintenance, no broken scripts at 2 AM.

Troubleshooting the three most common problems

"Server not found" after setup

If you're using the MCP server route, double-check the server URL in your Claude settings. For the skills route, make sure Node.js 18+ is installed. Run node --version to check. Node.js 22 LTS is the safest choice right now.

Transcripts returning empty

The video probably doesn't have captions. Check availability first with the captions or transcript-languages tool. Empty list? No transcript to pull.

Claude not auto-detecting YouTube links

For the MCP server setup, check two things. First, make sure TranscriptAPI is toggled on in your tools menu. Second, confirm the permission is set to "Allow unsupervised." Both need to be active for automatic detection.

If nothing else works, remove the connector entirely and add it again. That clears most persistent connection issues.

How to put this into action

1. Pick your setup path. MCP server for Claude.ai/Desktop, or YouTube skills for Claude Code and other agents

2. Follow the steps above. Either method takes under 10 minutes

3. Verify the connection by fetching a transcript from any YouTube video

4. Start simple. Summarize one video to see the output quality

5. Scale up to batch analysis, competitive research, or content repurposing once the workflow clicks

The bottom line

A YouTube MCP server turns Claude Code from a text-only tool into something that can search, browse, and analyze the world's largest video library. TranscriptAPI handles the infrastructure so you can focus on what matters: getting useful information out of video content fast.

What's the first YouTube video you'd throw at it?