- Home
- Skills
- Image & Video
- VideoToText | Local Transcription
More screenshots
Works with the AI Agents you already use
VideoToText | Local Transcription
Turn YouTube videos into timestamped transcripts, local summaries, and TXT, Markdown or JSON exports using whisper.cpp. Free and open-source.
Free
VideoToText | Local Transcription
Example session with this skill installed
Use VideoToText to transcribe https://youtu.be/jNQXAC9IVRw locally as structured JSON. Preserve timestamps and return the result without editing the recognized speech.
- Read your context and instructions
- Compiled the videototext
VideoToText — Local Transcription Result
Status: Success (Exit Code 0)
Source: https://youtu.be/jNQXAC9IVRw
Video: Me at the zoo
Model: whisper.cpp / ggml-base.en.bin
Output: Structured JSON
Segments: 4
The following is the complete, unmodified JSON output from a real Windows CLI transcription test.
{
"id": "job_6909f7ad-e195-4995-93b5-073edee694f3",
"url": "https://youtu.be/jNQXAC9IVRw",
"title": "Me at the zoo",
"createdAt": "2026-11-13T21:46:06.746Z",
"transcript": "[00:00:00 - 00:00:05] Alright, so here we are, one of the elephants.\n[00:00:05 - 00:00:13] Cool thing about these guys is that they have really, really, really long fums, and that's\n[00:00:13 - 00:00:17] cool.\n[00:00:17 - 00:00:18] And that's pretty much all I used to say.",
"summary": "Overview\nThis transcript covers \"Me at the zoo\" and runs for about 1 minute(s).\n4 spoken segment(s) were analyzed locally.\n\nKey points\n- Alright, so here we are, one of the elephants.\n- Cool thing about these guys is that they have really, really, really long fums, and that's cool.\n- And that's pretty much all I used to say.\n\nOpening\nAlright, so here we are, one of the elephants. Cool thing about these guys is that they have really, really, really long fums, and that's cool.",
"segments": [
{
"start": 0,
"end": 5.28,
"text": "Alright, so here we are, one of the elephants."
},
{
"start": 5.28,
"end": 13.42,
"text": "Cool thing about these guys is that they have really, really, really long fums, and that's"
},
{
"start": 13.42,
"end": 17.08,
"text": "cool."
},
{
"start": 17.08,
"end": 18.4,
"text": "And that's pretty much all I used to say."
}
]
}
Notes
- Transcription was performed locally using whisper.cpp.
- No paid transcription API was required.
- Timestamps are estimates produced by the transcription pipeline.
- Recognition errors are preserved in the original output.
- This demonstrates a verified CLI execution on Windows, not a completed Agensi MCP session.
Next steps
The transcript can also be exported as TXT or Markdown. A shell-capable AI agent can use the JSON segments to reference specific timestamps or create notes grounded in the recognized speech.
Connects securely to your tools. The creator never sees your data.
About this skill
Turn YouTube videos into timestamped transcripts and extractive summaries with local speech recognition.
VideoToText is a free, open-source transcription tool built with whisper.cpp, FFmpeg and yt-dlp.
What you can do
- Transcribe supported YouTube videos locally.
- Generate timestamped transcripts and extractive summaries.
- Export transcripts as TXT, Markdown or structured JSON.
- Ask your AI coding agent to analyze transcripts and reference timestamps.
- Process speech without a paid transcription API.
How it works
This skill instructs a shell-capable AI agent to use the separately installed VideoToText CLI. Audio is downloaded using yt-dlp, processed with FFmpeg, and transcribed locally using whisper.cpp.
Requirements
Install VideoToText, Node.js 22.13+, Git, Corepack/pnpm, yt-dlp, FFmpeg, whisper.cpp and a compatible GGML speech model separately.
Agensi provides the agent instructions, not the application or model. The skill cannot transcribe videos without a configured local execution environment.
Privacy and limitations
Speech recognition runs locally, but downloading YouTube audio requires an internet connection. Transcript text may be shared with your AI agent's provider.
Speech recognition can make mistakes. Summaries are extractive, timestamps are approximate, and visual content is not analyzed. Only process videos you have permission to download.
Price: Free ($0)
Licensing
VideoToText's source code and agent instructions are MIT-licensed and independently available on GitHub. When accessed through Agensi, delivery and use are additionally subject to Agensi's platform terms, including restrictions on saving or exporting skill content through its agent connection. These platform restrictions do not change the MIT license of copies obtained independently from GitHub.
Source and setup: https://github.com/serhataydilek/videototext
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Video Skills.
Requires local installation of yt-dlp, FFmpeg, and whisper.cpp. No support for private or age-gated videos. Extractive summary is text-only and ignores visual data.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 1 day ago
- Free with an Agensi account
- Software
Node.js
GitFFmpegyt-dlpwhisper.cpp
Read access to local tool configuration, installed model files and application files. Write access to temporary audio-processing files and user-approved transcript export destinations. No access to passwords or credentials required.
Permissions
Runs the separately installed VideoToText CLI and yt-dlp, FFmpeg/ffprobe and whisper-cli subprocesses. Reads installed application/tool files, the GGML model, local configuration and configured environment variables. Creates and deletes temporary downloaded/normalized audio and intermediate output; writes user-selected exports after checking for existing files. Network access downloads the selected video from YouTube/media hosts; authorized setup also uses GitHub, the npm registry and model-download hosts such as Hugging Face. No connected service account, credentials, paid API or remote inference is required. Transcript text may be sent to the user's AI agent provider.