vocce-transcribe-mcp
v1.1.0
Published
MCP server for vocce's hosted media tools: transcription, summaries, study notes, flashcards, document Q&A, and PDF compression. No API key required.
Maintainers
Readme
vocce media — MCP server
Transcribe, summarize, take notes, make flashcards, ask questions, and compress PDFs — straight from your AI agent. Powered by vocce's hosted backend (Whisper large-v3 + LLM tools).
✨ No API key · ☁️ Hosted (no model, no GPU to run) · 🆓 Free preview
Install
Add this to your MCP client config (Claude Desktop, Cursor, Windsurf, …). There's no install step — npx fetches it on first run:
{
"mcpServers": {
"vocce": {
"command": "npx",
"args": ["-y", "vocce-transcribe-mcp"]
}
}
}Restart your client and six tools are ready.
Tools
| Tool | What it does |
|---|---|
| transcribe | Local audio/video → text + SRT + VTT (100+ languages) |
| summarize | Recording or text → meeting minutes / lecture notes / to-dos / chapters |
| generate_notes | Recording or text → structured study notes (outline / cornell / key-points) |
| generate_flashcards | Recording or text → 5–20 grounded Q/A flashcards |
| ask_document | Ask questions about a file — answered only from the source, with verbatim citations |
| compress_pdf | Local PDF → smaller PDF (light / balanced / strong), text stays crisp |
For the content tools, pass either file_path (media is transcribed automatically first;
.txt/.md/.srt/.vtt are read directly) or raw text.
Example prompts:
"Transcribe
~/Downloads/interview.mp4and pull out the action items." "Make 15 flashcards from~/lectures/biology-week3.mp3." "Ask~/meetings/standup.mp4: what did we decide about the launch date?" "Compress~/Documents/report.pdfto email size."
Configuration
| Env var | Default | Description |
|---|---|---|
| VOCCE_API_BASE | https://api.vocce.io/api/v1 | Media backend (transcribe, PDF) |
| VOCCE_EDGE_BASE | https://vocce.io/api | LLM tools (summaries, notes, cards, Q&A) |
How it works
This server is a thin client: it uploads your file to vocce's hosted backend, polls until the job is done, and hands back the result. Heavy lifting runs on vocce's GPUs — you don't download or run a model locally.
Links
- Website — https://vocce.io
- Source — https://github.com/cmoonagent/vocce-agent
- Issues — https://github.com/cmoonagent/vocce-agent/issues
License
MIT
