Introduction

In the DeepSeek Harness (DSH) runtime, MCP server and skill library configurations are scattered across configuration files. This plugin lets you manage both centrally from the settings page, without manually editing YAML. The plugin maintainer, chihy525, focuses on solving the fragmentation of skill-package management and tool configuration.

Plugin Overview

This is a management panel located on the DSH settings page for managing MCP servers and the skill library.

Core Features

1. Manage MCP Servers

Open the MCP Servers panel under the MCP & Skills menu on the settings page:

  • List and Status: Displays all servers and supports enabling/disabling. HTTP servers automatically detect reachability, distinguishing plugin activation from the server’s actual running status.
  • Connection Testing: Supports stdio and streamable-http transports. Clicking Test performs a handshake and retrieves the tool list.
  • Tool Statistics and Estimation: After clicking Collect or Test, it snapshots the server’s tool list, and the card displays tools: N · ~X tokens. Tool counts and token estimates are based on community rules (approximately 4 bytes/token).
  • Configuration Items: Supports configuring authorization headers for streamable-http, and environment variables for stdio and http.

2. Manage Skill Library

Open the Skills panel under the MCP & Skills menu on the settings page:

  • Browse and View: Supports browsing standard skill root directories (.dsh/skills, .agents/skills, ~/.dsh/skills, ~/.agents/skills). Click View to see the full SKILL.md content.
  • Template Creation: Create skills from a template, supporting kebab-case naming, description, and the whenToUse field.
  • Grouped Views: Provides three views: by collection, by root, and flat. The default view groups by collection.
  • Batch Operations: Supports batch enabling/disabling model invocation, or deleting an entire collection.

3. Skill Grouping and Rules

Skills are grouped by source, either automatically detected or rule-based.

  • Rule Editor: Supports 10 matcher types (e.g., name regex, path substring, Frontmatter key-value, metadata deep path, etc.).
  • Suggestions and Testing: Provides a Feature Suggestions list, suggesting ungrouped skills. Supports Suggest (list then confirm to apply) and Auto (apply immediately) modes. Click Test to verify rule matching effects.

Typical Usage

Create a Skill

  1. Click Create on the skill library page.
  2. Select from a template and fill in a name (e.g., browser-skill) and description.
  3. Optionally fill in the whenToUse field.
  4. After saving, you can view it in the collection or flat view.

Configure an MCP Server

  1. Open the MCP Servers panel and click Add.
  2. Enter the server information and configure the transport.
  3. In the authorization header or environment variable area, add configuration in Key: Value format.
  4. Click Test Connection to verify.

Customize Grouping Rules

  1. Click “Grouping Rules” on the Skills page.
  2. Add a new rule and select a matcher type (e.g., hasKey).
  3. Switch between “Suggest / Auto” mode.
  4. Return to the Skills page, view “Feature suggestions,” select the skills, then click “Apply selected.”

Data Storage and Permissions

  • Dependency: Requires @deepseek-ai/cordis (^4.0.1).
  • Data Paths: Settings are saved in ~/.dsh/dsh-mcp-skill-manager/settings., skill data is saved in catalog., and collection data is saved in collections..
  • Permission Requirements: Write permission is required to create or edit configurations; the system automatically creates backups.

Use Cases

Suitable for users who need fine-grained management of MCP server connection status, tool-call cost estimation, and classification of a large number of skill packages (such as superpowers or hack-skills).

Conclusion

This plugin provides DSH users with a centralized entry point for MCP and skill management. Through visualized grouping and statistics, it reduces configuration complexity. For more details, refer to the GitHub repository.