MiniMax H3 MCP

Model Context Protocol server for MiniMax H3 multimodal video generation through AceDataCloud.

Capabilities

• Text-to-video

• Image-to-video with 1–9 reference images

• Audio-guided video with 1–3 audio references and optional images

• 4–15 second output in 16:9 or 9:16

• Asynchronous single and batch task retrieval

• Hosted OAuth HTTP transport and local stdio transport

Related MCP server: minimax-omni

Tools

Hosted server

https://minimax.mcp.acedata.cloud/mcp

The hosted server supports AceDataCloud OAuth. MCP clients that support remote OAuth can connect directly to this URL.

Public API reference: MiniMax H3 Videos API.

Local installation

pipx install mcp-minimax
export ACEDATACLOUD_API_TOKEN="YOUR_API_TOKEN"
mcp-minimax

Get a token from AceDataCloud.

Claude Code

claude mcp add minimax --transport stdio \
  --env ACEDATACLOUD_API_TOKEN=YOUR_API_TOKEN \
  -- mcp-minimax

Generic MCP configuration

{
  "mcpServers": {
    "minimax": {
      "command": "mcp-minimax",
      "env": {
        "ACEDATACLOUD_API_TOKEN": "YOUR_API_TOKEN"
      }
    }
  }
}

Examples

Text:

Generate a 6-second 16:9 video: a red fox running through a snowy forest at dawn, low tracking shot.

Images:

Animate these two reference images for 8 seconds while preserving the character and clothing.

Audio:

Create a 9:16 dance video guided by this audio, with cuts and motion following the beat.

Generation tools submit asynchronously. Poll the returned task ID with minimax_get_task until the final AceDataCloud CDN video is available.

Development

pip install -e ".[dev,test]"
ruff check .
pytest --cov=core --cov=tools
mypy core tools

License

MIT