@corepackai/mcp-server

The Official Model Context Protocol (MCP) Server for Corepack AI. This server bridges your Local AI Agents (Claude, Cursor, etc.) with your Corepack AI Context (.ai).

Note: This package is typically bundled with the CLI. You usually just want to run npm install -g corepackai.

Features

• Boot Protocol: One-shot initialization via corepackai keyword.
• Security: Restricted read/write access ONLY to the .ai directory.
• Context Awareness:
  • read_resource('corepack://local/active'): Reads Active Plan & Handoffs.
  • read_resource('corepack://local/graph'): Reads Knowledge Map.
• Tools:
  • boot_agent: Initializes session.
  • search_context: Wildcard/Regex search across all packs.
  • read_outline: Efficient large-file browsing.
  • read_file / write_file / patch_file: Safe file operations.

Usage

1. via Corepack AI CLI (Recommended)

This server comes included with the CLI.

npm install -g corepackai

Then configure Claude to use the CLI's bridge command: corepackai mcp.

2. Connect to Claude Desktop (Config)

Add to your claude_desktop_config.json:

{
  "mcpServers": {
    "corepackai": {
      "command": "corepackai",
      "args": ["mcp", "--cwd", "D:\\path\\to\\your\\repo"]
    }
  }
}

3. Ignition

Once connected, simply say:

"corepackai"

The Agent will load your project's context, verify its permissions, and await instructions.

Capabilities & Constraints

To prevent hallucinations, the Server enforces:

• NO access to source code (src/, package.json, etc) outside .ai.
• NO terminal command execution.
• ONLY file operations within .ai.