FunASR
Free工业级语音识别 MCP 服务，GPU 170 倍实时，支持 50+ 语言、说话人分离、情感检测。本地转写无需 API Key。
FreeFree tier
About FunASR
Model Context Protocol server that gives AI assistants local audio transcription with SenseVoiceSmall by default. Supports Docker deployment, MCP directory submission (Glama, Registry), and requires no API key for local processing.
Key Features
Local audio transcription via MCP protocol
Uses SenseVoiceSmall model by default
Docker container with stdio MCP interface
No API key required (fully local processing)
Supports MCP directory submission (Glama, official Registry)
Open-source and free
Pros & Cons
Pros
- Free and open-source
- Local processing ensures privacy (no API key needed)
- Easy Docker deployment for consistent environment
- Compatible with MCP ecosystem (Glama, Registry)
- Uses state-of-the-art SenseVoiceSmall model
Cons
- Limited to audio transcription only (no other MCP tools)
- No GPU acceleration mentioned in this example
- SenseVoiceSmall may not support all languages
- Requires Docker knowledge for deployment
Best For
Transcribing audio files for AI assistantsLocal speech-to-text for MCP-compatible applicationsOffline audio transcription with Docker deployment
FAQ
How to install FunASR MCP Server?
Install dependencies with `pip install funasr`. Alternatively, use the Docker image built from the examples/mcp_server directory.
How to run the server with Docker?
Build the image with `docker build -t funasr-mcp examples/mcp_server` and run with `docker run --rm -i -e FUNASR_DEVICE=cpu --mount type=bind,src=<audio_path>,dst=/audio,readonly --mount type=volume,src=funasr-mcp-cache,dst=/root/.cache/modelscope funasr-mcp`.
What model does the server use?
By default, it uses SenseVoiceSmall for audio transcription.
Does the server require an API key or internet connection?
No, processing is local and does not require an API key. The model is downloaded on first use via ModelScope.
How do I submit this server to MCP directories like Glama?
Use the examples/mcp_server folder as the Docker build context and set the server command to `python /app/funasr_mcp.py`. The expected MCP tool is `transcribe_audio`.
How to pass audio files to the server?
Mount a host directory to /audio read-only inside the container, then pass tool paths under /audio (e.g., /audio/meeting.wav).