Preface

Peers working on DSH plugins or agents often encounter a specific requirement: integrating content production stages such as topic research, copywriting, cover images, and short videos into Harness conversation flows. The conventional approach is to wrap external tools or set up an MCP service to handle permissions and callbacks. beav-creator-dsh provides an alternative path—it integrates directly via the Cordis service and Harness tools without using MCP, connecting Beav’s Xiaohongshu (RED/RedNote) workflows and social media creator workflows into DeepSeek Harness. Below is an introduction to the plugin’s positioning, features, and onboarding steps.

What is this

beav-creator-dsh is a community plugin maintained by Jamailar, currently version 0.1.5, with the connector part licensed under MIT. It brings Xiaohongshu (RED/RedNote) topic research and trend analysis, knowledge management, copywriting, cover and post images, audio, short videos, and sustainable multi-step content production into DeepSeek Harness.

Its integration method is: exposing Beav’s workspace, project, task, approval, and verified artifacts to Harness via Cordis service, Harness tools, background tasks, slash commands, session events, and web client contributions. The entire link does not use MCP.

Note: This is a community plugin, not an official DeepSeek product or endorsement; the plugin runs as trusted local Host code.

Core Features

  • Xiaohongshu (RED/RedNote) topic and trend research, knowledge management
  • Copywriting, cover and post images, audio, short videos
  • Sustainable multi-step content production
  • Long-running tasks are presented as native Harness jobs and replayable Beav task cards
  • Completion determination has hard validation: a task is only accepted when Beav reports persistent artifact read-back verification, and every returned artifact can be read back via the Creator Gateway
  • The connector cannot automatically approve paid or destructive operations; AI orchestration, knowledge access, media generation, approval, billing, persistence, and artifact verification belong to Beav

Installation and Enablement

Dependency requirements:

  • Beav 2.7.4 or higher
  • DeepSeek Harness 0.1.0-rc.6 or higher (rc.7 / rc.8 web clients require beav-creator-dsh@0.1.5 or higher)
  • Node.js 22 or higher (engines: node >=22.13)

Install the plugin to a Harness profile:

dsh plugin --profile web add beav-creator-dsh

After installation, start this profile, open Settings → Plugins → Beav Creator, select Connect Beav to complete authorization. Harness will open beav://connect/authorize, Beav will launch if needed and display the requested permissions; after approval, the connector completes a one-time, five-minute valid PKCE exchange, credentials are stored in BEAV_CREATOR_TOKEN. The token is not passed via clipboard, URL, browser client, chat, session events, or tool results.

If Beav is not running, operations requiring it will open beav://open and perform bounded health retries. You can override local endpoints or credential references in a profile patch:

- update:
    id: beav
    config:
      endpoint: http://127.0.0.1:41700
      tokenRef: BEAV_CREATOR_TOKEN
      autoLaunch: on-demand
      requestTimeoutMs: 30000

Note, the connector only accepts loopback HTTP endpoints.

Typical Usage

There are three common entry points.

The first is natural language requests, for example:

Use Beav to research a Xiaohongshu topic and turn the latest knowledge in my launch workspace into a post, cover images, and a 60-second video.

The second is typing @beav and selecting a stable workspace or project reference.

The third is slash commands, which execute operations directly without a model turn:

/beav connect
/beav status
/beav open
/beav workspaces
/beav new
/beav import
/beav save

Long-running tasks will appear in the form of native Harness jobs, accompanied by replayable Beav task cards. The determination of task completion is based on the artifact read-back verification mentioned above, ensuring that the generated content is genuinely persistent rather than just reporting completion within the conversation.

Use Cases and Considerations

Suitable for creators and agent developers who need to continuously produce content on social media platforms like Xiaohongshu but also want to put research, copywriting, and media generation into DSH sessions and task systems.

Before using, there are a few points to note:

  1. The plugin runs with the permissions of the current dsh process. It is trusted local Host code; it is recommended to check the source code and license before installation.
  2. The MIT license applies only to the connector, not to the Beav application or service. Beav’s desktop application, creator runtime, knowledge engine, media pipeline, and commercial services are proprietary and are not included in the repository or npm package.
  3. The connector only connects to Beav on loopback addresses. Requests contain bounded summaries and do not transmit raw media bytes through model tool parameters. Deep links contain a random request ID and PKCE challenge, but no token or verifier.
  4. Paid or destructive operations will not be automatically approved; approval rights reside on the Beav side.

Summary

beav-creator-dsh provides an MCP-independent integration path, connecting Beav’s Xiaohongshu topic research, copywriting, and multimodal content production capabilities to DeepSeek Harness’s tools, tasks, and session systems, and using artifact read-back verification to ensure the credibility of task completion.

Directory Page: https://www.skillhub.cn/plugins/Jamailar/beav-deepseek-harness
GitHub: https://github.com/Jamailar/beav-deepseek-harness