linkdigest-mcp

When you ask an AI assistant to read a social media link, it usually hits an app-download shell or a login wall. The real content lives inside videos and image notes, not in the HTML. linkdigest-mcp solves exactly this gap: it takes any social post URL and returns a plain-text digest a language model can actually use, complete with transcript, on-screen OCR, per-image descriptions, caption, author, publish date, and key points.

It runs as a hosted remote MCP server over streamable HTTP, so there is nothing to install or keep running locally. Hand it a post URL and you get back a structured summary; Mandarin speech is transcribed with SenseVoice, which is meaningfully more accurate on Chinese than Whisper-class models. The service has been validated against real posts on Xiaohongshu, Douyin, TikTok, YouTube, X, and ordinary web articles, while Bilibili and Facebook are explicitly marked as unsupported. Every response also carries a transcript_source field that distinguishes native captions from ASR output, and a degraded field that spells out in plain words what did not fully work, preventing a mis-transcribed number from being quoted back as fact.

If you use Claude Code, Cursor, or another AI tooling surface and need your assistant to digest social media content on the fly, or you build Dify workflows that require post text extraction, this plugin slots in directly. A REST API and Apify Store actors are also available for batch processing or integration into other automation pipelines.