Summary

The whole clipping pipeline, on your machine, drivable by an agent.

It is the rare tool where the agent surface is not an afterthought: transcription through to publishing is exposed as tools, and the knowledge base means the titles and picks come out in your show's voice rather than a generic one.

What it is

A local podcast clipper with three front ends: a CLI, a web studio, and an MCP server so a coding agent can do the whole thing through conversation. One command transcribes the episode, picks the moments worth clipping, crops to whoever is speaking, and burns the captions in — transcription and rendering happen on your machine.

What you get

26 MCP tools, enough for an agent to run transcription, scoring, rendering and publishing end to end
Clips in 9:16, 16:9 or 1:1 with captions sized per canvas, face tracking that follows the speaker, split-screen layouts, and multi-segment cuts that drop filler and long pauses
Four caption styles — branded, hormozi, karaoke, subtle — plus logos, intros, outros and background music from a reusable asset library, managed through `manage_assets`
Finding the moments: Whisper transcription with speaker diarization, or your own `.txt`, `.srt` or `.vtt`; AssemblyAI as an alternative engine; yt-dlp to pull an episode from a URL; audio energy and laughter detection for highlight reels
A knowledge base that teaches it your show's voice and title formulas, and checks suggestions against your episode database so it stops resuggesting moments you already published — edited in the studio or through `knowledge_base`
The studio at `localhost:3847` — library, episode workspace, per-clip detail, highlights, thumbnails, content, analytics, assets, knowledge and config, with a command palette across pages, clips and assets
Shipping — YouTube publishing with performance analytics, DaVinci Resolve export as FCPXML, presets, clip history with duplicate detection, and `manage_config` for the config profile

Requirements

No prerequisites to install: the shell or PowerShell installer fetches a self-contained binary, and the first run provisions Python, Node, FFmpeg, whisper.cpp and the models into a managed folder. Runs on macOS Apple Silicon, Linux x64 and arm64, and Windows x64. Clips land in `podcli-clips/` beside where you ran it unless `PODCLI_OUTPUT` says otherwise. Two settings matter for the agent path: `podcli mcp install` registers it with Claude Code, and `PODCLI_PORT` must be set in the MCP client's environment too, or the MCP tools render without live studio progress. Speaker diarization needs a Hugging Face token in `HF_TOKEN`. The only outbound calls are the optional AssemblyAI engine, YouTube publishing, and the optional AI clip scoring. AGPL-3.0, with a commercial licence available.

Setup effort

One command — curl -fsSL https://podcli.com/install.sh | sh