Hutash MCP server

Hutash ships an MCP server (hutash-mcp) that exposes your locally installed models as tools. Any MCP-compatible client — your MCP client, an MCP-compatible CLI, or other agents — can generate speech, music, voice clones, and transcriptions through your local models.

Hutash's own built-in assistant reaches your models through this same server and the CLI — it has no privileged path of its own. Anything it can do is reproducible as a command, and anything you can script is something you can hand to it.

Setup: Connecting a client

The engine must be running and the models you want must already be installed.

MCP-compatible CLI

One command registers the server:

<your-cli> mcp add hutash -- /path/to/hutash-mcp

Your MCP client

Add the server to your MCP client's config file:

{
  "mcpServers": {
    "hutash": {
      "command": "/path/to/hutash-mcp"
    }
  }
}

Restart the client after saving. The Hutash tools appear in its tool list.

Engine tools: Always available

Tool	Description
list_models	List installed models with status and capabilities
list_apps	List installed applications
show_model	Show a model's full detail and manifest
install	Install a model or app by ID
uninstall	Remove an installed model
start	Start a stopped model
stop	Stop a running model
get_status	Engine resource usage (RAM, VRAM, GPU)
get_activity	Recent install and lifecycle activity

Project tools: Writing into Studio

Tool	Description
list_projects	List all Studio projects
generate_in_project	Generate content and register it as a Studio asset in a named project — the same mechanism Studio's own UI uses

Per-model tools: One tool per installed capability

Tool names follow the pattern {model_id}_{capability}. They appear automatically when a model is installed and disappear when it is removed. Each tool's input schema is built from the model's own manifest, so there is no hardcoded per-model knowledge.

Tool	What it does
kokoro_tts	Generate speech with Kokoro
ace_step_1_5_music	Generate music with ACE-Step
chatterbox_clone	Clone a voice with Chatterbox
whisper_large_v3_stt	Transcribe with Whisper Large v3
qwen3_1_7b_improve	Refine text with Qwen3

Catch-all

generate_with_local_model runs inference against any installed model when the request doesn't match a specific per-model tool.

Requirements

• Hutash OS installed
• The engine (hutashd) running
• Models installed before their tools appear
• An MCP server restart to pick up tools from a freshly installed model

Vertical workflow tools: Subs and Podcast

Subs and Podcast are applications, not model pipelines. Once one is installed and running, each of its workflows is also available as an MCP tool — the same discoverability idea as the tools above, scoped to that app's own workflows. A tool's inputs and outputs match what its workflow declares, so an assistant sees the same fields a person filling in the app's own form would see.

Subs

Transcribe and translate video subtitles locally using AI models.

Tool	Description
transcribe	Transcribe — media_file (required), source_language. Returns: Subtitles (srt, vtt, txt); Raw Segments (json)
translate	Translate Subtitles — source_file (required), source_language, target_language (required). Returns: Translated Subtitles (srt, vtt, txt)
transcribe_translate	Transcribe + Translate — media_file (required), source_language, target_language (required). Returns: Original Subtitles (srt, vtt, txt); Translated Subtitles (srt, vtt, txt); Translated Segments (json)
render_styled	Render Styled Video — media_file (required), segments (required), style, platform, output_dir (required). Returns: Captioned Video (mp4); Styled Subtitles (ass)

Podcast

Turn podcast episodes into transcripts, show notes, and chapters.

Tool	Description
transcribe	Transcribe Episode — media_file (required). Returns: Transcript (srt, vtt, txt); Raw Segments (json)
full_process	Full Episode Processing — media_file (required). Returns: Transcript (srt, vtt, txt); Show Notes (txt); Chapters (txt)
notes_only	Show Notes from Audio — media_file (required). Returns: Show Notes (txt); Transcript (srt, vtt, txt)
chapters_only	Chapters from Audio — media_file (required). Returns: Chapters (txt); Transcript (srt, vtt, txt)
chapters	Generate Chapters — transcript (required). Returns: Chapters (txt)
show_notes	Generate Show Notes — transcript (required). Returns: Show Notes (txt)