> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kymaapi.com/llms.txt
> Use this file to discover all available pages before exploring further.

# watch-cli: give your agent eyes and ears

> Turn any video URL into frames plus a full transcript your agent can reason over. Transcription runs on Kyma; a one-hour video costs about $0.05.

[watch-cli](https://github.com/sonpiaz/watch-cli) composes `yt-dlp`, `ffmpeg` and a Whisper-class transcription model into one command. Your agent gets timestamped frames and the full transcript, then builds the artifact you asked for: a diagram, a component, a notebook, a cheat sheet. The only paid step is transcription, which runs through your Kyma key.

## Quick setup

<Steps>
  <Step title="Get your API key">
    Sign up at [kymaapi.com/signup](https://kymaapi.com/signup?src=docs:watch-cli). New accounts start with free credit on the [free tier](/pricing#free-tier), which covers `whisper-v3-turbo`.
  </Step>

  <Step title="Install watch-cli">
    ```bash theme={null}
    brew install sonpiaz/tap/watch-cli
    export KYMA_API_KEY=kyma-your-api-key
    ```

    Without Homebrew, run the release installer pinned to a version. It downloads the tagged tarball and checks its SHA256 against the checksum on the same GitHub Release:

    ```bash theme={null}
    curl -fsSL https://github.com/sonpiaz/watch-cli/releases/download/v0.3.4/install.sh | WATCH_CLI_VERSION=0.3.4 bash
    ```
  </Step>

  <Step title="Watch a video">
    ```bash theme={null}
    watch https://www.youtube.com/watch?v=dQw4w9WgXcQ
    ```

    The command prints the frame paths and the transcript in one block. Repeat watches are cache hits and cost nothing.
  </Step>
</Steps>

## As a skill or MCP server

* Skill for Claude Code, Cursor, Codex, Gemini CLI and others: `npx skills add sonpiaz/watch-cli`
* OpenClaw: `clawhub skill install watch-cli`
* MCP server: see the `mcp-server/` folder in the repository

## What leaves your machine

The video is downloaded from its platform; frames stay on disk. Only the audio track is sent to Kyma for transcription. Login-walled videos fail unless you opt in with `WATCH_BROWSER=auto`, which lets yt-dlp read cookies from a browser you are signed in to on the same machine and send them only to that platform. Nothing is read by default.

## Cost and models

Transcription uses Kyma's speech-to-text models, priced per minute of audio; see [Models](https://kymaapi.com/models?src=docs:watch-cli) for current rates. Frames are extracted locally and cost nothing.