Introduction

The core philosophy of DeepSeek Harness (DSH) is to modularize features as plugins to facilitate integration and extension. Processing specific business data locally is a common requirement in a plugin-based architecture. dsh-momo-learning is a plugin that syncs Maimemo vocabulary account learning data to a local SQLite database and performs local logic for question generation, incorrect-answer tracking, and daily report generation. The entire process is read-only with respect to the Maimemo account.

Core Features

This plugin mainly provides the following capabilities:

  • Local data mirroring: Syncs Maimemo account learning data to local SQLite.
  • Local question generation: Generates 20 multiple-choice questions daily based on local data.
  • Incorrect-answer management: Records the incorrect-answer book and accumulated incorrect answers.
  • Daily report generation: Generates a daily learning report.
  • Tools and Skill: Provides 19 MCP tools and one momo-daily-learning Skill.

Installation and Enabling

Prerequisites

  • System: Currently supports Windows only.
  • Node.js: Version must be >=24.13.1 <25 (keep it consistent with the Node runtime used by Harness).
  • DeepSeek Harness: The Harness environment must be installed.

Installation Steps

  1. Install MCP dependencies: Enter the plugin directory and install the dependencies required for the MCP service.
    cd dsh-momo-learning
    npm run mcp:install
  1. Add the plugin: Add the plugin to the specified profile.
    dsh plugin --profile <profile> add ./dsh-momo-learning
  1. Configure credentials: Credentials must not be written in configuration files; store them interactively in the Windows Credential Manager.
    npm run credential -- set
    npm run credential -- status

Usage Examples

MCP Tool Invocation

The plugin provides 19 MCP tools. For example, maimemo_list_quiz_candidates is used to retrieve candidate words:

maimemo_list_quiz_candidates({ limit: 20 })

Skill Interaction Example

After startup, local learning can be performed through the Skill. The following is an interaction example:

你：开始今天的

（弹窗，第 1/20 题）
  The committee will ___ the remaining issues at next week's meeting.
  ○ address    ○ dress    ○ redress    ○ regress

你选了 address

✓ 正确。address 这里是及物动词「处理、着手解决」，不是常见的名词「地址」。
  搭配：address an issue / a problem / a concern
  例句：The report addresses the risks of long-term debt.（这份报告论述了长期债务的风险。）
  易错点：后面接 issue / problem / question 时读作「处理」，跟「地址」没关系。

（第 2/20 题…）

20 题答完：答对 17 / 答错 3。错题进了本地错题本，明天会优先出它们。
日报写到 reports/momo-2026-09-14.md。

Notes and Limitations

  • Platform limitation: Currently supports Windows only; credentials are stored in the Windows Credential Manager.
  • Read/write permissions: Writing to the Maimemo account is not supported (adding words, changing review status, cloud word lists, etc.).
  • Configuration taking effect: After modifying SKILL.md, Harness must be restarted for changes to take effect.
  • Dependencies: The plugin declares inject: ['skills'], so the composition must include @deepseek-ai/dsh-skill.
  • MCP communication: The MCP service depends on stdio pipe support and may not work normally in restricted sandbox environments.

Short Conclusion

dsh-momo-learning provides a solution for managing Maimemo vocabulary data locally. By wrapping MCP tools and Skills, it integrates into the DeepSeek Harness workflow. It is suitable for users who need to process locally or use an LLM to generate definitions, explanations, and daily reports.