cast2md

Podcast transcription service -- download episodes via RSS and transcribe with Whisper. Automatically fetches publisher-provided transcripts (Podcasting 2.0) or Pocket Casts auto-generated transcripts before falling back to local transcription.

Note: This is a personal project under active development. I'm sharing it in case others find it useful, but I'm not currently providing support or reviewing pull requests.

Features

• Transcript-first workflow -- fetches external transcripts from Podcasting 2.0 tags and Pocket Casts before downloading audio
• Whisper transcription -- local transcription with faster-whisper or mlx-whisper (CPU, CUDA, Apple Silicon)
• Distributed transcription -- use remote machines (M4 Macs, GPU PCs) or RunPod GPU pods to transcribe in parallel
• Hybrid search -- full-text and semantic search across episode metadata and transcript content (pgvector)
• Web UI, CLI, REST API -- manage feeds, view episodes, search transcripts, monitor processing
• MCP server -- Claude integration via Model Context Protocol

Quick Start

git clone https://github.com/meltforce/cast2md.git
cd cast2md
cp compose.example.yml compose.yaml
cp .env.example .env
# Edit .env -- set POSTGRES_PASSWORD at minimum
docker compose up -d

Open http://localhost:8000 to access the web UI.

Which compose file. compose.example.yml is the template for running cast2md: it pulls the published image and requires POSTGRES_PASSWORD to be set. docker-compose.yml in this repo is the development stack -- it builds from source, hardcodes the database password to dev and publishes port 5432 to the host. Copying the example to compose.yaml as above takes precedence, so docker compose picks the right one.

Documentation

Full documentation is available at cast2md.meltforce.org.

Section	Description
Getting Started	Architecture and key concepts
Installation	Docker, manual install, transcriber nodes
Configuration	Environment variables, Whisper models
Usage	Web UI, CLI, REST API, MCP server
Distributed Transcription	Multi-machine setup, RunPod GPU workers
Deployment	Production deployment, server sizing

License

MIT