frank-ay/mimo-mcp

This page was translated automatically. Technical details may differ from the original. View the original

30 stars 2 forks Quality 70 Trend 15MIT

Packages Xiaomi MiMo's full-modal capabilities (conversation/image/video/TTS/voice cloning) into a stdio MCP Server, with Claude Code and Codex available for direct calling

Overview

The above image shows a 60-second preview (GIF, auto-loop); Full 108-second version with audio: 📥 mimo-mcp-intro.mp4 (3.2 MB) Xiaomi MiMo's full-modal capabilities—multimodal conversation / image / video understanding / TTS / voice cloning / sound design / ASR—are packaged into one, allowing Claude Code and Codex programming tools to be directly used as tool calls, accompanied by a ** (9 pages: overview / sandbox / text-to-speech / image video / sound library / cloning / design / speech transcription / audit). - Implemented in one place, Claude Code & Codex can be used - 11 MCP tools covering all modal capabilities from F1 to F8 - Local SQLite persistent voice library, create and design voices in one go for use - mimo.tts automatically routes voice types to corresponding MiMo models (default → tts / clone → voiceclone / design → voicedesign) - Includes Token Plan package adaptation + thinking model max_tokens Bottom-line pitfalls experience - Web console supports batch segmentation and batch compositing long texts by sentence end (SSE streaming rendering) - TTS supports v2.5 natural language style commands (director mode) and audio tags; ASR (mimo-v2.5-asr) bilingual Chinese-English + dialect recognition and automatic punctuation :./scripts/build_frontend.sh & ./scripts/run_web.sh — After one build, only the backend is run. FastAPI hosts dist, and you only need to access the address http://localhost:7801. When the backend is healthy, api_key_configured / base_url_reachable / auth_valid are all true. - Started with dev.sh or a single terminal: press Ctrl+C on that terminal (dev.sh will stop both the front and rear ends). - Started in the background: lsof -ti:7801,5173 | xargs kill You can also directly use uv run --directory /Users/Frank-ay/Desktop/xiaomi-MIMO mimo-mcp as command, then pass the MIMO_API_KEY in env.

Overview (machine translated)

The above image shows a 60-second preview (GIF, auto-loop); Full 108-second version with audio: 📥 mimo-mcp-intro.mp4(3.2 MB)

Xiaomi MiMo’s full-modal capabilities—multimodal conversation / image / video understanding / TTS / voice cloning / sound design / ASR—are packaged into a single stdio MCP Server, allowing Claude Code and Codex programming tools to be directly called as tools, accompanied by a local web management panel (9 pages: overview / sandbox / text-to-speech / image video / sound library / cloning / design / speech transcription / audit).

Highlights

Implemented in one place, both Claude Code and Codex can be used
11 MCP tools covering full-modal capabilities from F1 to F8
Local SQLite persistent voice library, allowing cloning/designing voices to be created and used everywhere at once
mimo.tts Automatically routes by voice type to the corresponding MiMo model (default → tts / clone → voiceclone / design → voicedesign)
Includes Token Plan package adaptation + thinking model max_tokens as a safety net and other pitfall experiences
Web console supports batch segmentation and batch compositing long texts by sentence ending punctuation (SSE streaming rendering)
TTS supports v2.5 natural language style commands (director mode) and audio tags; ASR (mimo-v2.5-asr) bilingual Chinese-English + dialect recognition with automatic punctuation

1. Three steps to get started

1.1 Installing dependencies

1.2 Configure API Keys

1.3 Startup

Fastest: One command to simultaneously pull up web frontend and backend (development mode), Ctrl+C stop together:

Open it and you’ll see the console.

More ways to start:

Purpose	Command
One-Click Development (Backend + Frontend, Recommended)	`./scripts/dev.sh`
stdio MCP server (for Claude Code / Codex)	`./scripts/run_mcp.sh`
Web console backend only (:7801)	`./scripts/run_web.sh`
Front-end development mode only (:5173, proxy to 7801)	`cd webui/frontend && pnpm dev`
Frontend built to dist (then only backend can be managed as a single process)	`./scripts/build_frontend.sh`

Production / Single Process: ./scripts/build_frontend.sh && ./scripts/run_web.sh — After one build, only the backend is run; FastAPI hosts dist, accessing only one address.

This page was translated automatically. Technical details may differ from the original.

Install

This server does not publish a one-line install command.

Open the repository installation guide New to MCP? Start with the mcpskill.org guide →

Configuration

{
  "mcpServers": {
    "mimo-mcp": {
      "command": "/Users/Frank-ay/Desktop/xiaomi-MIMO/scripts/run_mcp.sh"
    }
  }
}