A scraper that speaks fluent Claude

What it does

Feed it a Douyin share link and it spits back a watermark-free download URL plus an AI-transcribed transcript of the video’s audio. It wraps the whole pipeline — video fetch, audio extraction, speech-to-text via Silicon Flow’s SenseVoice API — behind a tidy interface.

The interesting bit

The MCP server angle. Instead of yet another standalone CLI tool, it exposes extract_douyin_text as a native tool to Claude Desktop, Cherry Studio, and friends. Your AI assistant can now ingest TikTok-style content directly during a conversation, which is either convenient or slightly dystopian depending on your workflow.

Key highlights

• Ships three access modes: browser-based WebUI, MCP server, and plain CLI
• Auto-splits audio over 1 hour / 50 MB into 9-minute chunks for API limits, then stitches transcripts back together
• Outputs Markdown with metadata tables and download links
• No API key needed just to grab video info or download links; transcription requires Silicon Flow key (free tier exists)
• “支持龙虾” in the repo description appears to be a garbled tag, not a feature

Caveats

• Hard dependency on external Chinese cloud API (Silicon Flow) for the speech recognition; no local fallback
• Disclaimer frames itself as “research use only,” which is prudent legalese for a watermark stripper

Verdict

Worth a look if you regularly need to quote or archive Douyin content and already live inside Claude Desktop. Everyone else can probably get by with yt-dlp and Whisper.