Turn a podcast episode into Shorts from Claude Code
Short answer
The commands below are copied from the Podcli README. Podcli was not reinstalled to write this page.
Nika Siradze · November 11, 2026
Commands
Same order as the README. Run them on your machine.
1. Install Podcli
On macOS or Linux run curl -fsSL https://podcli.com/install.sh | sh. On Windows PowerShell run irm https://podcli.com/install.ps1 | iex. The README says the installer fetches a self-contained binary. It runs on macOS (Apple Silicon), Linux (x64 and arm64), and Windows (x64). Intel Macs aren't supported.
2. Run setup once
Run podcli setup once before you point an agent at it. The README says podcli mcp does not provision the engine, because stdout is JSON-RPC. First run of process, studio, transcribe, or multicam provisions Python, Node, FFmpeg, whisper.cpp, and the models. --help and --version never download anything.
3. Register the MCP server
Run podcli mcp install. The README says this registers Podcli with Claude Code and Codex, whichever CLI is on PATH.
4. Clip the episode
Run podcli process episode.mp4 --json --yes, with the path to your file. The README says that command transcribes the episode, picks the moments worth clipping, crops to the visible face, and burns the captions in. --json writes one object to stdout and logs to stderr. --yes skips prompts. After the MCP server is installed, ask Claude Code to clip the same file.
5. Read the output
The README says clips land in podcli-clips/ in the directory you ran the command from. Run podcli doctor --json if a step fails. rank_clip_candidates returns a shortlist, and process_episode runs the pipeline as one job you poll with job_status.
Windows uses irm https://podcli.com/install.ps1 | iex instead of the curl line. Claude Desktop setup is in the MCP docs. The tool list is on the MCP page.
A short demo
The README links a demo with sound. Watch it here: the clip on X.
The home page plays /demo/hero.mp4 when that file is on the site. It is not in this repo, so the clip above is the one the README already links.
What the README says happens
Output can be 9:16, 16:9, or 1:1. Face tracking follows the visible face, and split-screen layouts are included. Following the active speaker needs AssemblyAI, or a source install with openai-whisper and HF_TOKEN. There are four caption styles.
The README also says the transcript is what leaves the machine for scoring: to podcli Pro if you are signed in, to Anthropic if ANTHROPIC_API_KEY is set, or to the provider behind your Claude Code or Codex login. Rendering stays local unless you turn on a cloud sync flag or a multicam cloud upload. Those flags are off unless you set them.
Other open source clippers, and what was actually installed for that comparison, are on open source OpusClip alternatives.
Questions
Answers taken from the same README.
Do I need an API key?
The README says clip scoring uses Claude Code or Codex login, or an Anthropic API key. Without those it falls back to heuristics. Transcription and rendering run on your machine.
Where do the files go?
Clips land in podcli-clips/ in the directory you ran the command from. The README says you can set PODCLI_OUTPUT to render somewhere fixed instead.
What caption styles ship?
The README names four: branded, hormozi, karaoke, and subtle.