transcribe.so MCP Server認証済み注目

What it does

transcribe.so turns a YouTube link, podcast page, direct media URL or uploaded file into a transcript with speaker labels and timestamps, then lets your AI assistant work with it:

• Who said what: search which speaker said something and when, across one recording or your whole library.
• Cited answers: ask a question about a recording and get an answer with timestamped citations.
• Chapters and timestamps: chapters, highlights and paste-ready timestamps for YouTube, podcasts and social posts.
• Subtitles: export SRT and VTT.
• Clips: render a captioned clip of a chosen moment.

Every result links back to the transcript in the transcribe.so dashboard, so you can check any claim at the exact second.

Connect

• Endpoint: https://transcribe.so/mcp (streamable HTTP)
• Auth: OAuth 2.0 with dynamic client registration and PKCE. Sign in with your transcribe.so account (email code or Google). An API key also works for headless agents.
• Claude: Settings > Connectors > Add custom connector > https://transcribe.so/mcp
• Claude Code: claude mcp add --transport http transcribe-so https://transcribe.so/mcp
• ChatGPT, Cursor, Codex, Gemini CLI: setup steps at https://transcribe.so/developers/docs#mcp

Example prompts

• "Transcribe this YouTube video and give me chapters with timestamps."
• "In my launch sync recording, who mentioned Rotterdam and when?"
• "Give me SRT subtitles for it."
• "Make a captioned clip of the part where they discuss pricing."

Notes

Tools that create or change something (transcribe, render a clip, delete) are marked as write or destructive, and the server asks the assistant to confirm before deleting. Chat clients cannot upload local files, so the assistant works from links; files are uploaded in the web app. Plugin, skill and CLI: https://github.com/shsunmoonlee/transcribe-agent