summarize_transcript

Instructions

Implementation Reference

• main.py:306-364 (handler)

  The main handler function for the summarize_transcript MCP tool. It extracts the video ID, fetches the transcript, and returns it structured with instructions, metadata, and transcript sections for LLM summarization.

  def summarize_transcript(
      url: str,
      prompt: str | None = None,
      languages: list[str] | None = None,
      include_timestamps: bool = False,
      include_metadata: bool = True,
  ) -> str:
      """Fetch a YouTube video's transcript and return it with summarization instructions.
  
      The LLM client should use the returned instructions and transcript to produce a summary.
      The output is structured into clearly-labeled sections so a human can review the prompt
      before letting the LLM act on it.
  
      Args:
          url: YouTube video URL or video ID
          prompt: Custom summarization instructions. If omitted, a default summary prompt is used.
          languages: Preferred languages in priority order (e.g. ["en", "de"]). Defaults to English.
          include_timestamps: When True, prefix each transcript line with [HH:MM:SS].
          include_metadata: When True (default), include a [VIDEO] block with title, channel, published date, duration, views, and description.
      """
      try:
          video_id = extract_video_id(url)
      except ValueError as e:
          return f"Error: {e}"
  
      langs = languages or ["en"]
      try:
          transcript = api.fetch(video_id, languages=langs)
          if include_timestamps:
              text = _format_transcript_with_timestamps(transcript)
          else:
              text = TextFormatter().format_transcript(transcript)
      except Exception as e:
          return _handle_transcript_error(e, video_id, langs)
  
      instructions = prompt or DEFAULT_SUMMARY_PROMPT
      prompt_source = "user-supplied" if prompt else "default"
      language = transcript.language
      language_code = transcript.language_code
      is_generated = transcript.is_generated
  
      sections = [
          f"[INSTRUCTIONS]\n{instructions}",
          f"[PROMPT_SOURCE]\n{prompt_source}",
      ]
  
      if include_metadata:
          meta = _fetch_metadata(video_id)
          sections.append(_format_metadata_block(meta, header="VIDEO"))
  
      sections.append(
          f"[METADATA]\n"
          f"Video ID: {video_id}\n"
          f"Language: {language} ({language_code})\n"
          f"Type: {'auto-generated' if is_generated else 'manual'}"
      )
      sections.append(f"[TRANSCRIPT]\n{text}")
  
      return "\n\n".join(sections)

• main.py:305-305 (registration)

  The @mcp.tool() decorator registers summarize_transcript as an MCP tool on the FastMCP server instance.

  @mcp.tool()

• main.py:25-28 (helper)

  Default summary prompt used when no custom prompt is provided by the user.

  DEFAULT_SUMMARY_PROMPT = (
      "Summarize the following YouTube video transcript. "
      "Provide a concise overview of the main topics, key points, and conclusions."
  )

• main.py:115-119 (helper)

  Helper function used to format transcript lines with timestamps (called when include_timestamps=True).

  def _format_transcript_with_timestamps(transcript) -> str:
      """Render a FetchedTranscript as text with an [HH:MM:SS] prefix on each line."""
      return "\n".join(
          f"{_format_timestamp(snippet.start)} {snippet.text}" for snippet in transcript
      )