MCP MCP server · io.github.modelscope
io.github.modelscope/funasr-mcp
Transcribe local audio with FunASR and SenseVoice using private, on-device inference.
- host stars
- 20,426
- 30-day movement
- +6923/day
- Related entries
- 1
- Connections
- 0
ocilocalstdio
An MCP server that exposes FunASR and SenseVoice speech recognition over stdio, transcribing local audio files with on-device inference. It builds on the FunASR toolkit, which lets you pick models (sensevoice, paraformer, fun-asr-nano), runtimes, and deployment targets independently.
You need private, local audio transcription without sending recordings to a cloud API.
Use it to
- Transcribe local audio files on-device
- Run SenseVoice as a CPU-only local service
- Deploy Fun-ASR-Nano with vLLM for batch processing
- Run ASR as a single binary on edge devices via llama.cpp/GGUF
- Add speaker-diarized long-form transcription with MOSS-Transcribe-Diarize
For Developers needing private, local speech-to-text
- Runs
- installed — on your machine
- Transport
- stdio
- Packaged as
- oci:ghcr.io/modelscope/funasr-mcp:0.1.2
- Install
- docker run ghcr.io/modelscope/funasr-mcp:0.1.2
- Version
- 0.1.2
- Last release
- 2026-07-26
topicsspeech-recognitionasraudio-transcriptionon-device-inferencesensevoicefunasr