What this is

MiniMax MCP is an OpenClaw skill bundle that turns multiple MiniMax capabilities into one coherent workflow.

Instead of manually picking separate tools every time, this project gives you:

• one skill definition for routing
• one set of local wrapper scripts
• one reference setup for OpenClaw
• one place to handle search, image analysis, and speech generation

It is especially useful if you want MiniMax inside OpenClaw to feel like one smart multimodal tool instead of three disconnected pieces.

Why it is useful

MiniMax MCP helps when you want to:

• search the web for fresh information
• inspect screenshots or images
• turn text into speech with official MiniMax Speech 2.8
• combine those steps into a single routed flow
• keep secrets in environment variables instead of scattered config files

Highlights

• Unified OpenClaw-facing MiniMax workflow
• Automatic routing between:
  • web search
  • image understanding
  • Speech 2.8 text to speech
• Simple local wrapper scripts for each capability
• Router example for combined multimodal flows
• Environment-variable based setup
• Compatible with uvx and local MCP tooling
• Good starting point for custom MiniMax automation

Quickstart in 30 seconds

1. Set your MiniMax credentials

export MINIMAX_API_KEY="your_key_here"
export MINIMAX_API_HOST="https://api.minimax.io"

2. Verify uvx

uvx --version

3. Try a capability

python scripts/minimax_web_search.py "latest MiniMax MCP updates"

4. Or let the router decide

python scripts/minimax_router.py "Explain this screenshot" --image ~/Desktop/error.png

Project structure

minimax-mcp/
├── README.md
├── SKILL.md
├── agents/
│   └── openai.yaml
├── references/
│   └── openclaw-setup.md
└── scripts/
    ├── minimax_router.py
    ├── minimax_tts.py
    ├── minimax_understand_image.py
    └── minimax_web_search.py

Core capabilities

Web search

Routes requests to the MiniMax MCP web_search tool.

Best for:

• current facts
• recent developments
• comparisons that need fresh data
• search-first research requests

Example:

python scripts/minimax_web_search.py "latest MiniMax MCP updates"

Image understanding

Routes requests to the MiniMax MCP understand_image tool.

Best for:

• screenshots
• UI analysis
• diagrams
• photos
• answering questions about visible content

Example:

python scripts/minimax_understand_image.py ~/Desktop/screenshot.png "What error is shown here?"

Speech 2.8 TTS

Uses the official MiniMax synchronous T2A HTTP endpoint and writes decoded audio locally.

Best for:

• narration
• spoken summaries
• voice output
• turning multimodal answers into audio files

Example:

python scripts/minimax_tts.py "Hallo zusammen" \
  --language German \
  --voice English_expressive_narrator \
  --format mp3 \
  --output ./hello.mp3

Smart routing

scripts/minimax_router.py is a simple orchestrator that chooses the most likely path from one request.

Examples:

python scripts/minimax_router.py "latest MiniMax MCP updates"
python scripts/minimax_router.py "Explain this screenshot" --image ~/Desktop/error.png
python scripts/minimax_router.py "Read this aloud" --speak --output ./speech.mp3

How routing works

The intended behavior is simple:

• need fresh information → web search
• need to inspect an image → image understanding
• need spoken output → TTS
• need multiple steps → chain them in sequence

That makes the skill practical for combined flows like:

• search, then speak
• inspect image, then explain as audio
• search first, then use the result in a follow-up voice output

File guide

SKILL.md

The OpenClaw-facing skill contract.

It defines:

• routing rules
• failure handling
• output expectations
• combined workflow behavior

references/openclaw-setup.md

Reference notes for:

• environment variables
• install assumptions
• MiniMax host configuration
• TTS defaults
• example commands

scripts/minimax_web_search.py

A thin wrapper that sends a tools/call payload to the MCP command for web_search.

scripts/minimax_understand_image.py

A thin wrapper for understand_image that validates the local image path and forwards the prompt.

scripts/minimax_tts.py

A direct HTTP wrapper for official MiniMax Speech 2.8 TTS.

It:

• reads MINIMAX_API_KEY from the environment
• sends a synchronous T2A request
• decodes the returned audio
• saves the output locally
• prints a structured JSON summary

scripts/minimax_router.py

A small local orchestrator example that auto-picks which MiniMax path to run.

Environment setup

Preferred environment:

export MINIMAX_API_KEY="your_key_here"
export MINIMAX_API_HOST="https://api.minimax.io"

Optional low-latency TTS host:

export MINIMAX_API_HOST="https://api-uw.minimax.io"

Recommended install location:

~/.openclaw/skills/minimax-mcp/

TTS defaults

The bundled TTS wrapper ships with sensible defaults:

• model: speech-2.8-hd
• voice_id: English_expressive_narrator
• language_boost: auto
• speed: 1.0
• volume: 1.0
• pitch: 0
• format: mp3
• sample_rate: 32000
• bitrate: 128000
• channel: 1

Typical requests this project supports

• "Search the latest MiniMax MCP news."
• "What is shown in this screenshot?"
• "Read this summary aloud in German."
• "Research this topic and save a spoken version as mp3."
• "Explain this UI screenshot as audio."

Credential check result

I checked the project structure for likely embedded credentials.

Current result:

• no obvious real credentials found
• MINIMAX_API_KEY appears only as:
  • environment variable examples
  • environment lookups in scripts
• no hardcoded bearer tokens, passwords, or real API keys were visible in the scanned files

That means the project currently looks clean from an obvious secret-leak perspective.

Security notes

• Prefer environment variables for secrets
• Do not hardcode MINIMAX_API_KEY in tracked files
• Avoid unreviewed curl | sh style install flows
• If you use local secret files, restrict file permissions
• If this project becomes public, keep example values clearly fake

Good fit for

This project is a nice fit if you want:

• one MiniMax skill instead of multiple separate OpenClaw skills
• a practical example of local MCP integration
• a base for richer multimodal routing
• official MiniMax TTS wrapped in a scriptable local interface