Hands-On Guide to Video Generation via MCP Tool Calling

Using the MiniMax MCP Server with Claude Desktop, Cursor, Windsurf & More

Large-language-model agents are far more useful when they can call real-world tools—and nothing shows that off better than turning a text prompt into a finished video. In this walkthrough you’ll learn how to:

• Spin up the official MiniMax Model Context Protocol (MCP) server
• Wire that server into popular MCP-enabled clients such as Claude Desktop and Cursor
• Generate speech, clone voices, create videos & images—securely and immutably

---

1 Prerequisites

What you need	Why you need it
MiniMax API key	Access to TTS, voice clone, video, image
uv package manager	Runs the Python-based MiniMax MCP server (uvx)
An MCP client	Claude Desktop, Cursor, Windsurf, OpenAI Agents…

# Install uv
curl -LsSf https://astral.sh/uv/install.sh | sh

# Install the MiniMax MCP server
uv pip install minimax-mcp

Region matters
Global → MINIMAX_API_HOST=https://api.minimax.io
Mainland → …minimaxi.com

---

2 Start the MiniMax MCP Server

export MINIMAX_API_KEY="sk-xxxxxxxx"
export MINIMAX_API_HOST="https://api.minimax.io"
export MINIMAX_MCP_BASE_PATH="$HOME/minimax-outputs"
export MINIMAX_API_RESOURCE_MODE="local"

uvx minimax-mcp -y

stdio vs. SSE

stdio (default)	SSE
Local only	Local or cloud
Uses stdout	Uses HTTP(S)
Accepts local files	Best with URLs

---

3 Add the Server to Your Client

<details> <summary>Claude Desktop JSON snippet</summary>

{
  "mcpServers": {
    "MiniMax": {
      "command": "/usr/local/bin/uvx",
      "args": ["minimax-mcp", "-y"],
      "env": {
        "MINIMAX_API_KEY": "sk-xxxxxxxx",
        "MINIMAX_API_HOST": "https://api.minimax.io",
        "MINIMAX_MCP_BASE_PATH": "/Users/you/Desktop/vids",
        "MINIMAX_API_RESOURCE_MODE": "local"
      }
    }
  }
}

</details>

Cursor IDE → Preferences → MCP → Add Global MCP Server (paste same block).

---

4 Available Tools

Tool	What it does
text_to_audio	Text → speech
list_voices	Fetch voice IDs
voice_clone	Clone a speaker
generate_video	Prompt → video
text_to_image	Prompt → image
query_video_generation	Poll async renders

---

5 Quick Examples

# List voices
mcp.tool list_voices

# Clone a voice
mcp.tool voice_clone src=['/path/alice.wav']

# Generate a teaser video
mcp.tool generate_video \
  prompt='A sleek electric bike racing across the desert at sunset' \
  duration=10s

:warning: Costs apply—set caps on your API key.

---

6 Async Mode

Define completion rules so your agent waits for query_video_generation to return status=success.

---

7 Workflows & Use-Cases

Scenario	Unlock
News teasers	Script → video → TTS
E-learning	Clone instructor voice & auto-render
Marketing	Storyboard → images → video
Accessibility	Multi-language dubbing
Game modding	NPC cut-scenes

---

8 Troubleshooting

• Invalid API key → host/key mismatch
• `spawn uvx ENOENT` → use absolute `uvx` path
• Slow queue → switch to async

---

9 Next Steps

Chain GPT-4o scripting → MiniMax video → auto-publish via an MCP postTweet() tool.

MCP + MiniMax turns text into production-ready multimedia—securely, immutably, and at scale.