mcp-youtube-transcript

What it does

This MCP server retrieves transcripts and metadata from YouTube videos via four tools. get_transcript returns plain-text transcripts; get_timed_transcript includes timestamps for each caption segment, enabling precise reference during editing or research. get_video_info fetches metadata like title, duration, and channel. get_available_languages lists available caption languages. For longer videos, transcripts exceeding 50,000 characters are split across multiple responses, with a next_cursor parameter for pagination. The server supports multiple caption languages and includes built-in handling for accessing YouTube from geographically restricted networks via proxy configuration.

Who it's for

Researchers analyzing video content for citations or evidence; content creators and analytics teams tracking video performance; developers building AI applications that need to extract video context; anyone accessing YouTube from networks with IP restrictions.

Common use cases

• Extract and analyze a full transcript from a YouTube video for research or fact-checking
• Retrieve timestamped transcripts for video editing, precise reference, or creating chapters
• Pull video metadata alongside transcript content for structured archival
• Access transcripts in languages other than English for multilingual research
• Fetch transcripts from geographically restricted networks using proxy servers

Setup pitfalls

• Long transcripts are paginated at 50,000 characters; always check for next_cursor and fetch continuation content to avoid incomplete results
• YouTube access may be blocked by IP; configure proxies via WEBSHARE_PROXY_USERNAME and WEBSHARE_PROXY_PASSWORD environment variables or HTTP_PROXY/HTTPS_PROXY
• Response size defaults to roughly 50k characters; use the --response-limit argument if your LLM has a smaller context window (e.g., --response-limit 15000 for 15k-char chunks)