youtube

YouTube Data MCP Server (@kirbah/mcp-youtube)

A production-grade YouTube Data MCP server engineered specifically for AI agents.

Unlike standard API wrappers that flood your LLM with redundant data, this server strips away YouTube's heavy payload bloat. It is designed to save you massive amounts of context window tokens, protect your daily API quotas via caching, and run reliably without breaking your workflows.

Why Choose This Server?

Most MCP servers are weekend projects. @kirbah/mcp-youtube is built for reliable, daily, cost-effective agentic workflows.

🎯 Want feedback on your own channel, not just raw data? Check out CreatorLens - a companion Claude Skill built on this MCP that diagnoses common growth problems (weak hooks, bad thumbnails, flatlined videos) using a real strategist's framework, not just numbers.

📉 1. Save Up to 87% on Tokens (and Context Window)

The raw YouTube API returns massive JSON payloads filled with nested eTags, redundant thumbnails, and localization data that LLMs don't need. This server structures the data to give your LLM exactly what it needs to reason, and nothing else.

%%{init: { "theme": "base", "themeVariables": { "xyChart": { "plotColorPalette": "#ef4444, #22c55e" } } } }%%
xychart-beta
    title "Token Consumption (Lower is Better)"
    x-axis ["getVideoDetails", "searchVideos", "getChannelStats"]
    y-axis "Context Tokens" 0 --> 1200
    bar "Raw YouTube API" [854, 1115, 673]
    bar "MCP-YouTube (Optimized)" [209, 402, 86]

(Curious? You can compare the raw API responses vs optimized outputs in the examples folder).

🛡️ 2. Protect Your API Quotas (Smart Caching)

The YouTube Data API has strict daily limits (10,000 quota units). If your LLM gets stuck in a loop or re-asks a question, standard servers will drain your API limit in minutes. This server includes an optional MongoDB caching layer. If your agent requests a video details or searches the same trending videos twice, the server serves it from the cache - costing you 0 API quota points.

🏗️ 3. Production-Grade & Actively Maintained

Tired of MCP tools crashing your AI client? This server is built to be a rock-solid dependency:

• 97% Test Coverage: Comprehensively unit-tested (check the Codecov badge).
• Zero Lint Errors/Warnings: Enforces strict, clean code (npm run lint passes 100%).
• Active Security: Automated Dependabot patching ensures underlying libraries are never left with known vulnerabilities.
• Strict Type Safety: Built using Zod validation and the robust MCP TypeScript Starter architecture.

Quick Start: Installation

🟢 Zero-Config Mode (No API Key Required)

Want to just fetch transcripts? You can use this server immediately without any configuration! Just install and go. Add a YouTube API key later to unlock deep search and analytics.

The easiest way to install this server is by clicking the "Add to Claude Desktop" button on the Glama server page.

If you are configuring manually (e.g., in Cursor), just add this bare-minimum setup:

{
  "mcpServers": {
    "youtube": {
      "command": "npx",
      "args": ["-y", "@kirbah/mcp-youtube"]
    }
  }
}

✨ Tip: In Zero-Config mode, you can ask your AI to simply "Read the transcript for youtube://transcript/{videoId}"!

🟡 Manual Configuration (Unlock All Features)

If you prefer to configure your MCP client manually (e.g., Claude Desktop or Cursor), add the following to your configuration file:

1. Get a YouTube Data API v3 Key (See Setup Instructions below).
2. (Highly Recommended) Get a free MongoDB Connection String to enable quota-saving caching.

{
  "mcpServers": {
    "youtube": {
      "command": "npx",
      "args": ["-y", "@kirbah/mcp-youtube"],
      "env": {
        "YOUTUBE_API_KEY": "YOUR_YOUTUBE_API_KEY_HERE",
        "MDB_MCP_CONNECTION_STRING": "mongodb+srv://user:pass@cluster0.abc.mongodb.net/youtube_niche_analysis"
      }
    }
  }
}

(Windows PowerShell Users: If npx fails, try using "command": "cmd" and "args": ["/k", "npx", "-y", "@kirbah/mcp-youtube"])

Next Step: Add a Skill

Once your server is connected, try CreatorLens - a Claude Skill built specifically for this MCP that turns raw YouTube data into growth diagnostics (weak hooks, packaging, niche doubt, format pivots).

Key Features

• Optimized Video Information: Search videos with advanced filters. Retrieve detailed metadata, statistics (views, likes, etc.), and content details, all structured for minimal token footprint.
• Efficient Transcript Management: Fetch video captions/subtitles with multi-language support, perfect for content analysis by LLMs.
• Insightful Channel Analysis: Get concise channel statistics (subscribers, views, video count) and discover a channel's top-performing videos without data bloat.
• Lean Trend Discovery: Find trending videos by region and category, and get lists of available video categories, optimized for quick AI processing.
• Structured for AI: All responses are designed to be easily parsable and immediately useful for language models.
• Efficient Comment Retrieval: Fetch video comments with fine-grained control over the number of results and replies, optimized for sentiment analysis and feedback extraction.

Available Tools

The server provides the following MCP tools, each designed to return token-optimized data:

For detailed input parameters and their descriptions, please refer to the inputSchema within each tool's configuration file in the src/tools/ directory (e.g., src/tools/video/getVideoDetails.ts).

Note on API Quota Costs: Most tools are highly efficient. getVideoDetails, getChannelStatistics, and getTrendingVideos cost only 1 unit per call. The getTranscripts tool has 0 API cost. The new getVideoComments tool has a variable cost: the base call is 1 unit, but if you request replies (by setting maxReplies > 0), it costs an additional 1 unit for each top-level comment it fetches replies for. The search-based tools are the most expensive: searchVideos costs 100 units and getChannelTopVideos costs 101 units.

Advanced Usage & Local Development