Foreword

The plugin ecosystem of DeepSeek Harness (DSH) expands through community plugins. dsh-mmx-bridge is one such MIT-licensed plugin, maintained by welsione. It provides DSH with access to MiniMax’s multimodal capabilities via an mmx_bridge tool, including image viewing, image generation, video generation, speaking, singing, audio cover, search, and quota checking.

Positioning

What Problem Does It Solve

dsh-mmx-bridge consolidates MiniMax’s multimodal invocations into a single DSH tool. This allows agents within DSH to request image understanding, image generation, video generation, speech synthesis, music generation, audio cover, web search, and quota querying during conversations.

Verified Capabilities

  • Image understanding (VLM)
  • Text-to-image generation
  • Text/image-to-video generation
  • Speech synthesis
  • Music generation
  • Audio cover
  • Web search
  • Quota query
  • Direct sending and automatic recognition after dragging in or pasting an image
  • Inline caching of image recognition results for reuse with the same image and question
  • Optional takeover to MiniMax/mmx-cli
  • Preview/playback of generated outputs in Web GUI, with a settings page management card

Tool and Interfaces

mmx_bridge is invoked by action:

describe
image
video
speech
music
cover
search
quota

The plugin optionally takes over the following invocation interfaces:

web_search
read_image

After takeover, these invocations are redirected to MiniMax/mmx-cli.

Installation and Enabling

Prerequisites

DSH 0.1.0-rc.7+
mmx-cli installed and logged in

Installation

dsh plugin --profile web add dsh-mmx-bridge

If npm fetch fails, use:

dsh plugin --profile web add github:welsione/dsh-mmx-bridge

Enabling

After installation, restart DSH and refresh the Web GUI.

Typical Usage

Conversation Examples

You can directly say in chat:

Describe this image
Generate an image of a cyberpunk cat
Turn this text into speech

Image Input

Drag in or paste an image and send normally. The Agent will automatically view the image. Recognition results form an inline cache, which can be reused when the same image and question are queried again.

Caching Note

The recognition cache is in plaintext. The cache becomes invalid if the image is re-encoded.

Configuration

Configuration can be adjusted via the “Plugin Configuration” settings page or control file. Listed configuration items:

enabled
count
webSearchEnabled
readImageEnabled
imageBridgeEnabled
imageCacheEnabled
mmxBin

Outputs and Access

Generated/bridged outputs are saved by default to:

/tmp/mmx-out/

And served via:

/mmx-files/

Security and Compatibility

  • License: MIT
  • Platforms: Supported on macOS/Linux; Windows is best-effort and not verified on real devices
  • External Dependencies: Requires mmx-cli to be installed and logged in
  • Data: No telemetry collection; no reading of system credentials; no modification of DSH attachment storage

Use Cases

Ideal for developers already using DSH who wish to integrate MiniMax’s multimodal capabilities into agent workflows. The plugin runs with the current DSH process privileges. Before installation, review the source code and license, and ensure mmx-cli is installed and logged in.

Closing

dsh-mmx-bridge encapsulates MiniMax’s multimodal capabilities as callable actions for DSH and provides preview, playback, and a settings page management card in the Web GUI. Repository address:

https://github.com/welsione/dsh-mmx-bridge