WhatsApp Voice Transcribe

Download WhatsApp voice notes, move them to the vault, transcribe them, and apply common mining site corrections.

What This Skill Does

1. Lists recent WhatsApp messages from a contact
2. Downloads audio messages
3. Moves files to vault media folder using Filesystem MCP
4. Transcribes using Whisper MCP with gpt-4o-mini-transcribe
5. Corrects common mining site mis-transcriptions

When to Use

• User says "transcribe my voice note"
• User asks to "check my WhatsApp voice messages"
• User mentions transcribing audio from WhatsApp
• User wants to process voice notes from a specific contact

Step-by-Step Workflow

Step 1: Find Voice Notes

whatsapp:list_messages
- chat_jid: 27833911315@s.whatsapp.net (or other contact)
- limit: 5
- include_context: false

Look for messages marked as [audio - Message ID: ... - Chat JID: ...]

Step 2: Download Media

whatsapp:download_media
- message_id: [from step 1]
- chat_jid: [from step 1]

File downloads to: C:\whatsapp-mcp\whatsapp-bridge\store\[chat_jid]\audio_[timestamp].ogg

Step 3: Move to Vault (CRITICAL)

Use Filesystem MCP, NOT bash cp:

Filesystem:move_file
- source: C:\whatsapp-mcp\whatsapp-bridge\store\[chat_jid]\audio_[timestamp].ogg
- destination: C:\Users\10064957\My Drive\GDVault\MarthaVault\media\audio\[descriptive_name].ogg

Why: Filesystem MCP operates on Windows directly. Bash runs in Linux container and cannot access C:\ paths.

Step 4: Transcribe

whisper:transcribe_audio
- input_file_name: [filename from step 3]
- model: gpt-4o-mini-transcribe
- response_format: text

Step 5: Apply Corrections

Common mining site mis-transcriptions to correct:

Example Usage

User: "Transcribe my latest voice note to myself"

Claude executes:

1. Lists messages from 27833911315@s.whatsapp.net
2. Finds latest audio: AC5B1B9A1C849D7E5803F94CAC9FB434
3. Downloads to WhatsApp store
4. Moves to media/audio/self_20251127_143637.ogg
5. Transcribes: "This is a test voice note..."
6. Applies corrections if needed
7. Returns corrected transcription

Common Contacts

• Self: 27833911315@s.whatsapp.net
• Jacques Breet: 27834418149@s.whatsapp.net
• Production Engineering Group: 27834418149-1537194373@g.us

Critical Technical Notes

Why Filesystem:move_file Works

• Desktop Chat runs in Linux container - bash commands can't see Windows paths
• Filesystem MCP runs on Windows - direct access to C:\ drives
• move_file handles binary files - no encoding issues
• Tested and working as of 2025-11-28

Why NOT to Use

❌ bash cp - Can't access Windows paths from Linux container ❌ read_media_file then write_file - Returns "unsupported format" ❌ Python shutil in bash - Same container limitation

✅ Filesystem:move_file - Works perfectly

Validation Checks

Before returning results, verify:

• ✓ Audio file downloaded successfully
• ✓ File moved to media/audio/ folder
• ✓ Transcription generated
• ✓ Common corrections applied
• ✓ Output formatted clearly

Output Format

Return structured output:

Transcription: "[corrected text]"
File: media/audio/[filename].ogg
Duration: [if available]
Corrections Applied: [list any corrections made]

Troubleshooting

"File not found" on transcription

• Verify Step 3 completed (file moved to media/audio/)
• Check filename matches exactly

"No such file or directory" with bash

• Don't use bash for file operations
• Use Filesystem:move_file instead

No audio messages found

• Verify contact phone number format
• Check WhatsApp MCP is connected
• Try with chat_jid instead of sender_phone_number

Integration Points

After transcription, consider:

• Extract action items → create tasks
• Add to master_task_list.md
• Link to relevant projects
• Create meeting notes if applicable

Status

✅ Production Ready - Tested 2025-11-28

• Workflow validated in Claude Desktop
• All MCP connections working
• Mining corrections dictionary loaded