Transcribe a meeting recording into text or Markdown on your own computer.
Generate SRT or WebVTT subtitle files from a video without using cloud services.
Let Claude Desktop or other AI tools transcribe audio files directly via an MCP server.
Transcribe audio on machines without a GPU using the Parakeet backend.
| sm18lr88/stt-mcp | 0-bingwu-0/live-interpreter | 010zx00x1/faresnipe | |
|---|---|---|---|
| Stars | 2 | 2 | 2 |
| Language | Python | Python | Python |
| Setup difficulty | moderate | moderate | easy |
| Complexity | 3/5 | 2/5 | 2/5 |
| Audience | vibe coder | general | general |
Figures from each repo's GitHub metadata at analysis time.
Requires installing uv and FFmpeg beforehand, plus choosing the correct speech recognition backend based on your hardware.
STT-MCP is a tool that lets AI assistants transcribe audio and video files into text. It works entirely on your own computer, so your media never gets sent to a cloud service. The software takes an audio or video file, converts it into a clean format using a tool called FFmpeg, runs speech recognition to produce a written transcript, and then removes any temporary files it created along the way. There are two speech recognition engines, called backends, and the one you use depends on your hardware. Parakeet is designed for computers without dedicated graphics cards, covering Windows, Linux, and Mac systems including Apple Silicon. Granite is built for machines with a specific NVIDIA graphics card and aims for the highest quality output on that hardware. Once you pick a backend, that choice sticks for both the command line tool and the server mode. Installation is guided by a detailed setup document. The developers recommend telling your coding assistant to read that document, inspect your computer, and recommend the right backend before installing anything. You need a Python tool called uv, plus FFmpeg, installed beforehand. After setup, you can transcribe files from the command line, choosing from several output formats: plain text, Markdown, JSON, SRT subtitles, and WebVTT subtitles. You can also run it as an MCP server, which means other AI tools like Claude Desktop or Opencode can call the transcription function directly. The software is designed to be careful. It only runs one transcription at a time, so concurrent requests are rejected rather than queued. If a transcription fails, it avoids leaving behind partial or broken output files. It never downloads model files during regular operation, only during explicit setup. The project does not currently declare a license, so you should not redistribute the source until one is added. The repository also skips automated cloud testing because proper verification depends on your specific hardware, your chosen backend, and a real audio sample of someone speaking.
A tool that lets AI assistants transcribe audio and video files into text entirely on your own computer, without sending media to the cloud.
Mainly Python. The stack also includes Python, FFmpeg, MCP.
No license is declared, so you can view the code but should not use, copy, or redistribute it until the authors add one.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly vibe coder.
This repo across BitVibe Labs
double-check against the repo, no cap.