English | 简体中文

Agent-Skills-Manager-MCP (MCP Server)

A Model Context Protocol (MCP) server that acts as an AI agent "skills library manager". It provides tools to manage and share AI rules/skills across various AI-enhanced IDEs and agent frameworks.

Features

• Query: Read from a central library of AI skills. Supports both flat .md files and Standard Agent Skills Format (directories containing SKILL.md with standard YAML metadata).
• Push: Extract IDE-specific rule files (like .cursor/rules/something.mdc with frontmatter) and save them to the central skill library precisely adopting the standard <skill_name>/SKILL.md format.
• Pull (Project): Pull a designated skill from the central library and format it for the specific environment in the current project (e.g., injecting YAML frontmatter natively).
• Pull (Global): Pull a designated skill and format it natively into the user's global tool rules directory.
• Delete (Opt-In): Delete a skill permanently from the central library. By default, this destructive tool is disabled to prevent accidental data loss. It must be explicitly enabled by passing the AGENT_SKILLS_ENABLE_DELETE=true environment variable to the MCP server.

Configurable Paths (ide-config.json)

Yes, this configuration file is fully functional and actively controls where skills are saved!

The server dynamically manages destination paths per AI tool. Upon first run, it generates a data/ide-config.json next to your skills library.

[!TIP] On Path Accuracy: Due to version updates and environmental differences among various tools, the default generated paths may not be 100% accurate. You are encouraged to manually adjust the paths in ide-config.json to match your actual toolchain layout.

Example structure:

{
  "cursor": {
    "project": {
      "rulesLocation": ".cursor/rules",
      "skillsLocation": ".cursor/skills"
    },
    "global": {
      "rulesLocation": "~/.cursor/rules",
      "skillsLocation": "~/.cursor/skills"
    }
  },
  "my-custom-cli": {
    "project": {
      "rulesLocation": ".mycli/rules",
      "skillsLocation": ".mycli/agents"
    },
...
}

How it works:

1. Dynamic Resolution: Whenever an AI tool requests to pull a skill (using the pull_skill_to_project or pull_skill_to_global tool), it passes its ide_name. The MCP server looks up that name in ide-config.json.
2. Project Locations: Paths under project are resolved relative to your current workspace root (e.g., .cursor/skills becomes /path/to/your/project/.cursor/skills).
3. Global Locations: Paths under global that start with ~/ are automatically resolved to your operating system's absolute user home directory (e.g., ~/.aider/skills seamlessly becomes C:\Users\Username\.aider\skills on Windows or /Users/User/.aider/skills on Mac).
4. Custom Tools: You can add any brand new CLI tool or AI editor simply by creating a new key in the JSON file. The MCP server will instantly support pulling skills into it.
5. Tool Variants: Different AI assistants or editors may have different path requirements. You can easily adapt the target locations by editing ide-config.json to match your local setup. The server will seamlessly route files to the correct folders even if the assistant uses a generic identifier.

⚠️ Note on "Rules" Management: While ide-config.json includes entries for rulesLocation alongside skillsLocation, this tool is primarily optimized for Standard Skills (directories containing a SKILL.md and auxiliary files synced to skillsLocation). The synchronization and formatting logic for disjoint flat rules files remains experimental and incomplete.

Build and Run

Prerequisites

• Node.js >= 20
• npm

Installation

git clone https://github.com/MTQwQ/agent-skills-manager-mcp.git
cd agent-skills-manager-mcp
npm install

Development

npm run dev

Build

npm run build

Then you can start the built server using:

npm start

Note: Depending on your environment, you might need to point the MCP client to the exact path of the script using node /absolute/path/to/agent-skills-manager-mcp/build/index.js.

Configuration Example

In your editor's MCP settings, add a new stdio connection:

{
  "mcpServers": {
    "agent-skills-manager-mcp": {
      "command": "node",
      "args": [
        "/absolute/path/to/agent-skills-manager-mcp/build/index.js"
      ]
    }
  }
}

Enabling the Delete Tool (Optional)

If you must use the delete_skill_from_library tool, you must inject the opt-in environment flag into your IDE connection settings like this:

{
  "mcpServers": {
    "agent-skills-manager-mcp": {
      "command": "node",
      "args": [
        "/absolute/path/to/agent-skills-manager-mcp/build/index.js"
      ],
      "env": {
        "AGENT_SKILLS_ENABLE_DELETE": "true"
      }
    }
  }
}

Other Tools (Example)

For Windsurf, you might define it in your mcp.json or equivalent configuration:

{
  "mcpServers": {
    "agent-skills-manager-mcp": {
      "command": "node",
      "args": [
        "/absolute/path/to/agent-skills-manager-mcp/build/index.js"
      ]
    }
  }
}

(Remember to replace /absolute/path/to/agent-skills-manager-mcp with the actual location on your file system.)

Data Directory

By default, the server manages a data/skills/ directory alongside the project's root. Skills are stored there in the Standard Agent Skills Format (each skill is a directory containing a SKILL.md file and other related technical assets).