Anthropic Skill Core (MCP Server)

A robust, lightweight Model Context Protocol (MCP) server designed to empower your AI Agents with context-awareness, safe execution sandboxes, and dedicated thought logs. Built on Anthropic Engineering best practices.

🌟 Capabilities

This MCP Server registers 4 specialized tools for any connected LLM Agent:

1. ping: Basic health check to ensure the MCP server is alive.

2. think: A sandbox for Chain-of-Thought (CoT). Agents can dump their thoughts here before acting. It automatically records these thoughts to a local logs/ directory for developer debugging.

3. contextual_search: A robust directory search tool. It returns not just the matching text, but the contextual window (surrounding lines) of the codebase, preventing the "lost context" issue found in simple RAG.

4. audited_run: A CLI execution sandbox tailored for Auto-Mode. It blocks destructive commands (rm, sudo, chown etc.) and enforces a rigid 30-second timeout, allowing agents to execute tasks (e.g. npm run build) autonomously and safely.

Related MCP server: Sandbox MCP

🚀 Installation (For Claude Desktop)

To install this MCP server into Claude Desktop, modify your configuration file (claude_desktop_config.json):

{
  "mcpServers": {
    "anthropic-skill-core": {
      "command": "node",
      "args": ["/absolute/path/to/anthropic_skill/dist/index.js"]
    }
  }
}

🛠️ Build & Contribution

git clone https://github.com/williamchek/anthropic_skill.git
cd anthropic_skill
npm install
npm run build

📜 Author & License

williamchek Available under the MIT License.