apple-podcast-plaud 0.1.0a0

Apple Podcasts → Plaud → text. Detect a downloaded episode, auto-upload to Plaud, wait for transcription, hand back to any agent.

apple-podcast-plaud

Apple Podcasts → Plaud → text. Detect a downloaded episode, auto‑upload to Plaud, wait for transcription, hand the transcript back to any coding agent.

What it does

You download a podcast episode in Apple Podcasts on your Mac. Then you ask your coding agent ("Claude Code", "Cursor", "Cline", or anything else) to transcribe it. The skill / CLI:

1. Locates the downloaded m4a from the Apple Podcasts SQLite + cache
2. Auto-uploads it to your Plaud account (Pro subscription required — uses your existing minute quota; the package itself has no transcription cost)
3. Polls until Plaud finishes transcription + AI summary
4. Writes a verbatim transcript Markdown, an AI summary Markdown, and a metadata JSON to a known location
5. Prints a JSON envelope to stdout — your agent picks it up and decides what to do next (write an article, ingest into a knowledge base, search, …)

For English podcasts that have a native Apple-generated transcript, the package can skip Plaud entirely and use the free Apple TTML cache instead (no quota burn, instant). See docs/language-routing.md.

Why

• Apple Podcasts does not generate native transcripts for Chinese (and many other) podcasts. Plaud Pro is great at Chinese ASR and offers AI summaries.
• Existing Plaud reverse-engineered tools (TypeScript, Python) don't fit a clean Apple-Podcasts-anchored agent workflow:
  • sergivalverde/plaud-toolkit — TS, no upload, only us/eu regions hardcoded
  • arbuzmell/plaud-api — Python, has upload but no Apple Podcasts integration, no APAC region support either
• This package bridges the two and is agent‑agnostic: it just produces a transcript file and tells you the path. Your agent decides downstream.

Status

Alpha. APIs may change. PyPI release pending. See CHANGELOG.md.

Install

pip install apple-podcast-plaud  # once published

For local dev (recommended — works around a Python 3.14 .pth bug we hit in setuptools' editable install on some Macs):

git clone https://github.com/jumpin-dev/apple-podcast-plaud
cd apple-podcast-plaud
./scripts/dev-install.sh
source .venv/bin/activate
apb --version

That script creates a venv, installs the package + dev extras, and writes both a plain-text .pth AND a PYTHONPATH export to .venv/bin/activate so import apple_podcast_plaud always works regardless of whether your Python build honours .pth files at startup.

Quickstart

1. One-time auth — pick one of two paths

You need a Plaud bearer token. There are two ways to get one — pick whichever works for your account.

Path A — email + password login (only works if your account has a password)

apb auth login                   # interactive: prompts for email + password
# or
apb auth login --email [e-mail]

Some accounts (Google sign-in only, no password ever set) cannot use this path; the server will reject every credential pair until you set a password via the "Forgot password" flow at https://web.plaud.ai. If you hit wrong account or password even with the right credentials, switch to Path B.

Path B — paste a token from the web app

Works for every account, including Google-only.

1. Log in to https://web.plaud.ai
2. Open DevTools → Console (NOT the Local Storage view — that pane line-wraps long values and copies will silently include newlines)
3. Run:

   copy(localStorage.getItem('tokenstr'))
   

4. Save the token:

   apb auth set-token             # interactive: paste then Ctrl-D
   

   …or do it by hand:

   mkdir -p ~/.config/plaud && pbpaste > ~/.config/plaud/token && chmod 600 ~/.config/plaud/token
   

Verify

apb auth status
# region: apac
# expires: 2027-02-19 (in 299 days)
# server check: ok

Tokens last ~10 months; rotate via the same flow when expired. Detail: docs/token-extraction.md.

2. Run

# Transcribe an episode you've already downloaded in Apple Podcasts.
# Match by keyword in podcast or episode title:
apb transcribe "逆商"

# Override output dir:
apb transcribe "逆商" --out-dir ~/Desktop/podcast-transcripts

# JSON envelope to stdout — your agent reads this:
apb transcribe "逆商" --json-out -

3. As an MCP server (recommended for Claude Code)

Add this to your Claude Code settings.json:

{
  "mcpServers": {
    "podcast-transcribe": {
      "command": "uvx",
      "args": ["apple-podcast-plaud", "mcp"]
    }
  }
}

Then just talk to Claude naturally: "帮我转写最新一期播客" or "list my downloaded podcasts". The MCP server exposes 4 tools:

Tool	Description
list_podcasts	Query downloaded episodes by keyword
transcribe	Upload + transcribe an episode via Plaud (2-5 min)
auth_status	Check if Plaud credentials are configured
auth_login	Log in with email + password

For local development, use:

{
  "mcpServers": {
    "podcast-transcribe": {
      "command": "/path/to/apple-podcast-plaud/.venv/bin/python",
      "args": ["-m", "apple_podcast_plaud.mcp"]
    }
  }
}

4. As a Claude Code skill (legacy)

Drop the claude-skill/ folder into ~/.claude/skills/apple-podcast-plaud/, or symlink it. See claude-skill/SKILL.md.

JSON envelope (output schema)

{
  "status": "ok",
  "podcast": "高情商沟通话术：自在表达，想说就说",
  "episode": "320 逆商：我们该如何应对坏事件？",
  "language": "zh",
  "duration_sec": 1815,
  "segment_count": 48,
  "out_dir": "/Users/.../Documents/podcasts/2026-04-26-逆商",
  "files": {
    "transcript_md": ".../transcript.md",
    "summary_md": ".../summary.md",
    "raw_json": ".../transcript.raw.json",
    "metadata_json": ".../metadata.json"
  },
  "source": "plaud",
  "plaud_recording_id": "e4eb71b...",
  "elapsed_sec": 187
}

Full schema: docs/output-format.md.

Architecture

src/apple_podcast_plaud/
├── plaud/      # Reusable Plaud API client (us/eu/apac region routing,
│               # auth, recordings, transcriptions). Could spin out as
│               # a separate package in the future.
├── bridge/     # Apple-Podcasts ↔ Plaud orchestration: SQLite querying,
│               # language detection, language-routed tracks (Plaud / Apple
│               # TTML), output writers, CLI.
└── mcp/        # MCP server for Claude Code integration — thin wrapper
                # over plaud/ and bridge/ modules.

Acknowledgements

This project is independently implemented but the API surface and request shapes were initially scouted from the excellent reverse-engineering work in:

• arbuzmell/plaud-api — MIT-licensed Python Plaud client; the upload pipeline and browser-headers profile here were studied from this project before being reimplemented.
• sergivalverde/plaud-toolkit — MIT-licensed TypeScript Plaud client; the region-redirect detection pattern was studied from this project.

Apple Podcasts SQLite/TTML schema is documented across:

• mattdanielmurphy/apple-podcast-transcript-extractor
• cvonste2/apple-podcast-transcript-tool
• dado3212/apple-podcast-transcripts

All trademarks belong to their respective owners. This project is not affiliated with or endorsed by Plaud Inc. or Apple Inc. The Plaud API used here is reverse-engineered from the public web app and may change without notice.

License

MIT. See LICENSE.