YouTube Transcripts, Search & Summaries

What YouTube Transcripts, Search & Summaries does

YouTube Transcripts, Search & Summaries MCP server gives an MCP client read-only access to public YouTube video information. It can retrieve a transcript with title, channel, duration, views, publication date, category, available caption languages, and whether the selected caption track is automatic. Transcript output can be plain text, timestamped paragraphs, structured segments, or SRT subtitles.

The server also searches YouTube, returns video and channel details, and supports filters for upload date, duration, sorting, subtitles, live videos, HD, and Creative Commons. Chapter results use creator-provided chapters or YouTube markers when available. When those are absent, the server can derive topic segments from the transcript.

Its summary function is extractive rather than generative. Results can include a TL;DR, de-duplicated points, keywords, chapter bullets, and notable quotes. Items include timestamps and links to the relevant moment in the video. These capabilities place it in the 🔎 Search & Data Extraction category.

How it works

The four MCP tools are get_transcript, search_videos, get_video_chapters, and summarize_transcript. The video argument accepts an 11-character YouTube ID or supported URLs from watch, youtu.be, Shorts, embed, live, mobile, music, and no-cookie hosts.

get_transcript selects captions using a prioritized language list. It can permit automatically generated tracks, fall back to another available language, request YouTube machine translation, limit output size, and choose the response format. search_videos uses YouTube's InnerTube search API and can retrieve additional pages up to the requested limit of 50.

get_video_chapters can prefer automatic selection, description chapters, or transcript-derived segments. summarize_transcript accepts limits for points, chapter bullets, quotes, and keywords. No LLM runs on the server, and all tools are read-only.

Setup and configuration

The repository can run locally with Node.js using npm install, npm run build, and node dist/index.js, which starts a stdio server. A hosted deployment is also available through an Apify Streamable HTTP endpoint. Hosted requests use an Apify API token in the Authorization: Bearer header.

MCP clients named in the documentation include Claude Desktop, Claude Code, Cursor, VS Code, and Windsurf. Clients with only stdio support can connect to the hosted endpoint through mcp-remote. The local command does not require a YouTube API key or Google account.

Tools and capabilities

• Retrieve captions in selected languages, including SRT and timestamped formats.
• Search videos with date, duration, sort, language-region, and feature filters.
• Return metadata such as views, duration, thumbnails, descriptions, live status, and badges.
• Extract creator chapters or infer topic segments from transcript content.
• Produce timestamped extractive summaries, keywords, chapter bullets, and quotes.
• Use YouTube machine translation when a suitable caption language is unavailable.

Limitations and notes

The server works with public, playable videos that expose captions when transcript features are requested. Private, removed, age-restricted, members-only, region-blocked, upcoming, or otherwise unplayable videos can fail. Captions may be disabled, unavailable in requested languages, empty, or blocked by YouTube's request controls. Translation also depends on YouTube supporting the requested target language.

YouTube may return request-blocked or token-related errors, including rate limits, bot checks, and caption URLs requiring a PO token. Tool failures are returned as MCP errors with a code, message, retryability flag, and optional details. Derived chapters are estimates based on transcript content, not creator-authored chapter markers. The hosted Apify version charges per successful tool call; pricing is separate from the local open-source setup and is not specified here.

YouTube Transcripts, Search & Summaries MCP server is best suited to research, citation, subtitle extraction, and video navigation where read-only access is sufficient. It does not modify videos, captions, or YouTube accounts.