How to Use Claude YouTube MCP | Subtitle Retrieval and Video Summarization Server

What Is Claude YouTube MCP — Bridging Claude and YouTube

Model Context Protocol (MCP) is a common protocol published by Anthropic that connects Claude (Desktop / Code / API) with external tools and data sources. YouTube MCP servers fall into three main types: "reading" servers that retrieve subtitles (transcript / caption), "data" servers that call the YouTube Data API v3 for video search, channel statistics, and comment retrieval, and "hybrid" servers that integrate both.

Claude on its own cannot directly watch or analyze YouTube videos, but by pulling in subtitle text via an MCP server, tasks like summarization, quotation, and chapter structuring become Claude's strong suit. Compressing a 10-minute tutorial into 1,000 words, extracting only a specific speaker's lines from an hour-long conversation, or cross-referencing multiple videos and comparing them chapter by chapter — all of this can be done entirely within Claude's chat, without returning to the browser.

Official specification: https://modelcontextprotocol.io/

Recommended Servers by Use Case — Choosing Between Subtitle, Data, and Hybrid Types

There are currently over 10 YouTube MCP servers published on GitHub, each serving a different role. The following table organizes the major implementations for easy comparison.

※ Star counts are approximate as of May 2026. MCP-related projects update quickly — always check the latest star counts and package names in each repository's README.

The type classification is straightforward: subtitle (transcript) readers are suited for summarizing or quoting a single video in depth; Data API servers are suited for cross-referencing many videos through search, statistics, or comments; hybrid servers are suited when you want to handle both web page summarization and subtitle retrieval through a single server. Supplementary notes on each server follow.

1. kimtaeyoon83/mcp-server-youtube-transcript (TypeScript): The most widely used subtitle-specialized server — just pass a video URL and it retrieves the full subtitles. Supports multiple languages including Japanese and English. https://github.com/kimtaeyoon83/mcp-server-youtube-transcript
2. jkawamoto/mcp-youtube-transcript (Python): Also subtitle-specialized, running entirely within a Python environment. Ideal for those who prefer virtual environments or use uv on Mac. https://github.com/jkawamoto/mcp-youtube-transcript
3. anaisbetts/mcp-youtube (JavaScript): A lightweight video info retrieval server. Short codebase makes it easy to read and understand — good for beginners. https://github.com/anaisbetts/mcp-youtube
4. ZubeidHendricks/youtube-mcp-server (TypeScript): A full-featured server wrapping the YouTube Data API v3, supporting search, statistics, comments, and playlist operations. https://github.com/ZubeidHendricks/youtube-mcp-server
5. egoist/fetch-mcp (TypeScript): General-purpose URL fetch with YouTube subtitle extraction bundled in. Convenient when you also want web page summarization. https://github.com/egoist/fetch-mcp

The selection guideline is simple: for use cases like podcasts and lectures where you want to "read one video deeply," go with a subtitle-specialized server; for research or competitor channel analysis where you need to "survey many videos at once," go with a Data API server; if you need both, registering two servers simultaneously is the practical solution.

Setting Up the Subtitle Server — Installing the kimtaeyoon83 Version in Claude Desktop

The steps to install the most widely adopted mcp-server-youtube-transcript into Claude Desktop are as follows. First, make sure Node.js 18 or later is installed.

1. Run npx -y @kimtaeyoon83/mcp-server-youtube-transcript --help in your terminal to confirm the package can be fetched.
2. Open ~/Library/Application Support/Claude/claude_desktop_config.json on macOS, or %APPDATA%\Claude\claude_desktop_config.json on Windows.
3. Inside the mcpServers key, add an entry named youtube-transcript with command: "npx" and args: ["-y", "@kimtaeyoon83/mcp-server-youtube-transcript"].
4. Fully quit Claude Desktop and restart it, then confirm that youtube-transcript appears in the plug icon at the bottom right of the input field.
5. Paste a video URL into the chat and ask "Get the subtitles for this video and summarize it in 500 characters" to verify that the subtitles are being loaded.

If it does not work, check the Claude Desktop logs (~/Library/Logs/Claude/mcp*.log) — the MCP server's stderr is logged there. Look first for Cannot find module or 403 errors. Installation details: https://github.com/kimtaeyoon83/mcp-server-youtube-transcript#installation

Setting Up the YouTube Data API Server — Running the ZubeidHendricks Version with an API Key

For search, statistics, and comment retrieval, ZubeidHendricks/youtube-mcp-server is currently the most feature-complete option. This requires a YouTube Data API v3 key from Google Cloud.

1. Create a new project in Google Cloud Console, then go to "APIs & Services" → "Library" and enable YouTube Data API v3.
2. Generate an API key under "Credentials," and apply HTTP referrer or IP restrictions if needed.
3. Install globally with npm install -g youtube-mcp-server, or run it on demand with npx -y youtube-mcp-server.
4. Add an entry named youtube-data to the mcpServers section of claude_desktop_config.json, and set YOUTUBE_API_KEY: "..." under env.
5. Restart Claude Desktop and verify it works by asking something like "List the titles and view counts of the 5 most recent videos from the official Anthropic channel."

The YouTube Data API v3 free quota is 10,000 units per day, and each search consumes 100 units, so it can be exhausted quickly depending on usage. For commercial use or ongoing research, applying for a quota increase or combining it with the subtitle MCP to reduce API calls is the standard approach. Repository: https://github.com/ZubeidHendricks/youtube-mcp-server

Usage Scenarios — Summarization, Comparison, Comment Analysis, and Channel Monitoring

Combining subtitle retrieval and the Data API allows the following practical tasks to be handled entirely within Claude's chat window.

• Chapter-structured summaries of long videos: Point the subtitle MCP at an hour-long conversation or keynote and ask "Divide this into 5 chapters, summarize each in 200 characters, and include one key quote per chapter" — you can get all the way to a blog draft in one shot.
• Cross-video comparison: Load the top 5 videos on the same topic via subtitles and ask Claude to output "claims made consistently across all videos," "conflicting claims," and "unique perspectives not found in other videos" in a matrix format to significantly improve research quality.
• Comment sentiment analysis: Retrieve top comments with the Data API server, then hand them to Claude for positive/negative classification and frequent topic extraction — a viewer insight survey done in 10 minutes.
• Channel monitoring dashboard: List the latest videos from a specific channel on a weekly basis and generate subtitle summaries in one pass — a system to ensure you never miss a video from creators you follow.
• Creating study materials and quote collections: Extract remarks made just before code appears on screen from technical tutorial videos and format them as Markdown to create searchable, reusable notes.

To improve summarization quality, pass the subtitles to Claude with line breaks preserved and instruct "keep timestamps next to each section heading." Summaries with timestamps make it much easier to jump back to specific parts of the video, significantly improving reading efficiency.

Common Pitfalls and Solutions — No Subtitles, Region Locks, Token Limits

Here is a summary of the most common issues encountered during setup and how to work around them.

• Subtitles returned as "none": Some channels have auto-generated subtitles disabled, or subtitles may not yet be available for recently published videos. Wait 10–30 minutes and retry, or specify subtitles in another language (e.g., auto-generated English).
• 403 / region locked error: Videos with geographic restrictions imposed by rights holders cannot be retrieved server-side. There is no workaround — the practical solution is to find a different video source. Bypassing via VPN falls outside the intended use of MCP servers and is not recommended.
• Subtitles too long and exceeding context: A two-hour conversation can produce tens of thousands of subtitle tokens. Switch to Claude Sonnet 4.6's 1M token mode, or implement a pipeline that chunks the subtitles before merging the summaries.
• Command not found / won't start: This is often a PATH resolution issue with npx. Replacing command in claude_desktop_config.json with the full path (e.g., /opt/homebrew/bin/npx) instead of just npx usually fixes it.
• API quota exhausted: Search-related calls to the YouTube Data API v3 are heavy and the daily limit is easy to hit during development. Caching search results to a local JSON file via a companion MCP, or switching to a subtitle-first workflow, are effective solutions.

Official troubleshooting: https://developers.google.com/youtube/v3/getting-started

Conclusion — The Dual-Wielding Approach of Subtitle + Data API Is the Practical Answer

The most practical setup for Claude YouTube MCP is a two-layer approach: use a subtitle-specialized server (mcp-server-youtube-transcript / mcp-youtube-transcript) to make video content "readable," and a Data API server (youtube-mcp-server, etc.) to make video metadata "searchable." The recommended starting configuration is to register both servers side by side in Claude Desktop's config file and handle all three use cases — summarization, analysis, and quote collection — within a single chat.

MCP server updates move fast, and features or package names can change. Before installing, check each repository's README and release history, and version-control the diff in your claude_desktop_config.json so that troubleshooting "it suddenly stopped working" is much easier.

Reference: https://www.anthropic.com/news/model-context-protocol