Vectis MCP Server

Vectis is a powerful Model Context Protocol (MCP) server that connects AI agents to the skills.sh directory. It enables agents to search, download, and execute expert blueprints and instructions directly within their workflow.

🚀 Features

• Expert Blueprints: Access over 4000+ specialized skills from skills.sh.
• Fuzzy Search & Ranking: Advanced scoring system with popularity (leaderboard) support.
• Sandboxed Execution: Installs skills in isolated temporary directories for safety.
• Detailed Telemetry: Comprehensive logging of all tool interactions and internal logic.
• Auto-Cleanup: Automatically purges temporary files after execution to prevent disk bloat.

🛠️ Installation

Quick Start

git clone https://github.com/xenitV1/vectis.git
cd vectis
npm install
npm run build

Prerequisites

• Node.js (v18+)
• skills-cli (npm install -g @skills/cli)

Configuration

Add Vectis to your MCP settings (e.g., Claude Desktop or Cursor):

{
  "mcpServers": {
    "vectis": {
      "command": "node",
      "args": [
        "[ABSOLUTE_PATH_TO_VECTIS]/dist/index.js"
      ]
    }
  }
}

> [!IMPORTANT]
> Replace `[ABSOLUTE_PATH_TO_VECTIS]` with the actual full path where you cloned the repository on your machine.

## 📜 Execution Protocol (Gold Standard)

For the best results, the AI agent should always follow this protocol:

1. **Search First**: Always use `search_skills` to find the most relevant and popular blueprint.
2. **Download & Audit**: Use `execute_skill` to fetch instructions.
3. **Check References**: If the skill details refer to other skills or external docs, follow them.
4. **Script Retention**: If the skill contains **SCRIPTS** or complex automation logic:
   - Set `keepSandbox: true`.
   - Keep the skill available until the project is 100% finished.
   - Manually call `clear_cache` only after the entire project is completed.

🔧 Tools

search_skills

Search for professional blueprints and expert instructions.

• Input: query (string)
• Output: List of skills with descriptions and repo URLs, ranked by relevance and popularity.

execute_skill

Fetches the full expert instructions for a specific skill.

• Input: repoUrl, skillName, keepSandbox (optional)
• Output: Full content of SKILL.md from the requested blueprint.

clear_cache

Purges all temporary sandbox directories.

🧪 Development

1. Clone the repository
2. Install dependencies: npm install
3. Build the project: npm run build
4. Run in dev mode: npm run dev

🛡️ License

This project is licensed under the MIT License - see the LICENSE file for details.

📈 Star History

👤 Author

Developed by xenitV1. Follow me on X: @xenit_v0