A public video, as text
Paste a link to a public TikTok, Reel, YouTube or Facebook video and get a timestamped transcript you can check, copy, and hand to ChatGPT, Claude or Cursor. Numbers stay as spoken. Unclear words are flagged, not guessed.
Or upload a file (MP4, MOV, WebM, M4A, MP3 or WAV). Uploads are always available as the fallback.
Stop scrolling. This camera setting costs nothing and changes every photo you take.
Open the menu, go to Picture Style, and set it to Neutral.
Then shoot at ISO 400 for a cleaner night shot.
The Lumix S5Low confidence The speech model was unsure of these words. Check them against the audio before you quote them. does this in one tap. The older model needs two.
Save the preset, and you're done.
How it works
Paste
Public YouTube, Shorts, TikTok, Instagram Reel and Facebook video links work. Uploads are always available as the fallback, within the size limit.
Transcribe
Platform captions are used when they exist. Otherwise speech recognition runs, and every line keeps its timestamp.
Use
Copy it, export SRT, VTT, Markdown, TXT or JSON, or ask ChatGPT, Claude or Cursor to summarize, translate or pull the hooks.
Fidelity
Analysis is only as good as the text under it. We change formatting, never meaning, and we say so when we are not sure.
What the speech model returned
so the kit is $10, and you need the Lumix S5 firmware before you start.
What MuteScroll returns
So the kit is $10, and you need the Lumix S5Low confidence The speech model was unsure of these words. Check them against the audio before you quote them. firmware before you start. (Low confidence. Check the camera name against the audio before you quote it.)
Changed
Kept exactly
We tidy spacing and the first capital letter, nothing else. We never rewrite words, and if a number would differ after cleanup, we keep the original line, so $10 never becomes $100.
Words the speech model was not confident about are highlighted, with the reason. Check them against the audio before you quote them.
Voices are labelled Speaker 1, Speaker 2. We never guess who someone is, and we never add names we cannot see in the audio.
Use it where you work
MuteScroll also runs as an MCP server. Add it to your assistant once, then ask in plain language. The assistant fetches the transcript itself and works from the text.
Turn on Developer Mode in ChatGPT settings, then add a custom MCP connector with the address below. Then ask: “Transcribe this: [link]”.
Open Claude's connector settings, add a custom connector, and paste the address below. Then ask Claude to transcribe a link.
Add this JSON to Cursor's MCP settings and enable the server. Then ask the agent to transcribe a link and summarize it.
{
"mcpServers": {
"mutescroll": {
"url": "https://api.mutescroll.xyz/mcp"
}
}
}MCP server address
https://api.mutescroll.xyz/mcp
Hosted assistants need a public https address. Tools: transcribe_video and get_transcription.
Questions
YouTube and Shorts: public videos, using captions when they exist. Facebook: public videos. TikTok and Instagram: public posts on a best-effort basis. Retrieval is often blocked, so uploading the file is the reliable route. Upload is always available as the fallback, within the size limit. We never use logins, cookies or workarounds to reach private or restricted content.
No. We fetch only the audio needed for the transcript, never return the media file, and delete temporary media within 24 hours.
The link, the transcript, and basic details such as title, author and duration. Anonymous results are kept for 24 hours. Uploads are never shared between users.
No. We don't use your media or transcripts to train models.
We don't promise perfect accuracy. Speech recognition makes mistakes, especially with names and noisy audio. We flag the words the model was unsure of, and we keep numbers exactly as spoken. Check names and numbers before you publish.
Anonymous use has an allowance of 5 credits per rolling 24 hours. One credit covers up to 10 minutes of audio. Failed jobs never use credits. If a file is too large, the message shows the upload limit.
No. MuteScroll returns the text with timestamps. Your assistant (ChatGPT, Claude or Cursor) does the summarizing, translating and analysis, using the transcript in your conversation.
You are. Only process content you have the right to use, and remember that each platform's terms still apply. For takedown requests, see the privacy page.