YouTube Comment Downloader MCP Server

A Model Context Protocol (MCP) server that provides AI systems with the ability to download and analyze YouTube video comments without requiring API keys.

Features

• 4 specialized tools for different comment analysis needs
• No authentication required - uses web scraping
• Context-efficient statistics tool to avoid token bloat
• Built-in capacity planning with memory and timeout limits
• Engagement analysis with actual like-count sorting

MCP Client Configuration

Add this configuration block to your MCP client (e.g., Claude Desktop):

"ytcomment-mcp": {
  "command": "uv",
  "args": [
    "run",
    "--directory",
    "/Users/chad.kunsman/Documents/PythonProject/ytcomment_mcp",
    "src/server.py"
  ]
}

Available Tools

1. download_youtube_comments

Download raw comment data with full details.

• Parameters: video_id, limit (1-10000), sort (0=popular, 1=recent)
• Returns: Full comment dataset with all metadata
• Use case: When you need complete comment data for analysis

2. get_comment_stats

Get statistical analysis without full comment data (context-efficient).

• Parameters: video_id, limit, sort
• Returns: Statistics + 5 sample comments (~200 tokens vs ~25,000)
• Use case: Quick engagement insights without context bloat
• Triggers: "how engaged", "what's the engagement", "comment patterns"

3. search_comments

Search for specific terms within comments.

• Parameters: video_id, search_term, limit, sort
• Returns: Matching comments + search metadata
• Use case: Finding mentions, sentiment analysis, topic research
• Triggers: "find comments about", "search for", "mentions of"

4. get_top_comments_by_likes

Get most-liked comments sorted by actual like count (not YouTube's "popular").

• Parameters: video_id, top_count (1-100), sample_size (100-2000, default: 500)
• Returns: Top comments ranked by likes + engagement stats
• Use case: Finding viral comments that YouTube's algorithm might not surface first
• Triggers: "most popular", "most liked", "viral comments", "best comments"

Quick Start

# Install dependencies
uv venv && source .venv/bin/activate
uv pip install -e .

# Test functionality
python test_server.py

# Run MCP server
python src/server.py

Data Structure

Each comment contains 11 fields:

• cid, text, time, time_parsed, author, channel
• votes (likes), replies, photo, heart, reply

Capacity: ~1.8KB memory, ~25 tokens per comment

Key Limitations & Performance

• Flat structure: No hierarchical reply threading
• Mixed results: Top-level + replies mixed together (~10%/90% split)
• Rate limited: Built-in delays, ~30-90 sec per 500-1,000 comments
• Timeout handling: Larger requests may timeout; tool includes fallbacks
• No API quotas: Web scraping approach, but respect YouTube's terms

Performance Optimizations

• Reduced timeouts: 90s default (was 120s) for faster failure detection
• Smaller defaults: 500 comment samples (was 1000) for better reliability
• Timeout fallbacks: get_top_comments_by_likes tries recent sort if popular fails
• Context efficiency: Stats tool uses ~200 tokens vs ~25,000 for full data

Example Usage

# Get engagement overview (context-efficient)
stats = await get_comment_stats("dQw4w9WgXcQ", limit=1000)

# Find specific mentions
results = await search_comments("dQw4w9WgXcQ", "rickroll", limit=500) 

# Get viral comments by actual likes
top = await get_top_comments_by_likes("dQw4w9WgXcQ", top_count=20)

Built with FastMCP and youtube-comment-downloader.