name: mcp-server description: Model Context Protocol server exposing AI Factory tools to Claude Desktop version: 1.0.0 type: mcp-server author: AI Factory

AI Factory MCP Server

Revolutionary AI-native integration - Exposes your entire AI Factory through the Model Context Protocol, enabling Claude Desktop to directly access skills, files, and tools.

🎯 What This Does

This MCP server creates a native bridge between Claude Desktop and your AI Factory, giving Claude:

✅ Direct access to all 45+ skills
✅ Read/write file operations
✅ Project management
✅ Context saving/loading
✅ Analytics access
✅ Search capabilities

🏗️ Architecture

┌──────────────────┐
│  Claude Desktop  │ 
│   (macOS/Win)    │
└────────┬─────────┘
         │ MCP Protocol (stdio)
         ↓
┌──────────────────┐
│  This MCP Server │ ← You are here!
│  (Node.js)       │
└────────┬─────────┘
         │ Direct file access
         ↓
┌──────────────────┐
│   AI Factory     │
│  ~/.claude-skills│
│  ~/.ai-factory   │
│  ~/projects      │
└──────────────────┘

📋 Available Tools

Skills Management

skills_list - List all available skills with descriptions
skill_read - Read complete skill content (SKILL.md or script)
skill_execute - Execute read-only skill commands

File Operations

file_read - Read any file from AI Factory directories
file_write - Write files to .ai-factory/ directory
directory_list - List directory contents

Context Management

context_save - Save project context (JSON)
context_load - Load saved project context

Project Operations

project_list - List all projects
project_init - Initialize new project

Analytics & Favorites

analytics_get - Get usage analytics
favorites_list - List favorite skills
favorites_add - Add skill to favorites

Search

search_skills - Search for skills by keyword

🚀 Installation

Step 1: Install Dependencies

cd ~/.claude-skills/core/mcp-server.skill
npm install

Step 2: Configure Claude Desktop

On macOS:

Edit ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "ai-factory": {
      "command": "node",
      "args": [
        "/home/alex/.claude-skills/core/mcp-server.skill/scripts/start.js"
      ]
    }
  }
}

On Windows:

Edit %APPDATA%\Claude\claude_desktop_config.json:

{
  "mcpServers": {
    "ai-factory": {
      "command": "node",
      "args": [
        "\\\\wsl.localhost\\Ubuntu\\home\\alex\\.claude-skills\\core\\mcp-server.skill\\scripts\\start.js"
      ]
    }
  }
}

Step 3: Restart Claude Desktop

Close and reopen Claude Desktop. The MCP server will start automatically.

💡 Usage Examples

List All Skills

You: "List all my AI Factory skills"

Claude Desktop: 
[Calls skills_list tool]
"You have 45 skills available:
- task-manager: Personal task and TODO management
- time-tracker: Track time spent on projects
- bridge: Context sharing between Claude instances
..."

Read a Skill

You: "Show me the task-manager skill"

Claude Desktop:
[Calls skill_read with skill_name: "task-manager"]
"Here's the task-manager skill:
[displays complete SKILL.md content]"

Execute a Skill

You: "What tasks do I have today?"

Claude Desktop:
[Calls skill_execute with skill_name: "task-manager", command: "list today"]
"You have 3 tasks for today:
1. [HIGH] Complete documentation
2. [MED] Review code
3. [LOW] Update changelog"

Save a Design

You: "Save this design as feature-spec.md"

Claude Desktop:
[Calls file_write with path and content]
"Design saved to ~/.ai-factory/designs/feature-spec.md"

Search for Skills

You: "Find skills related to testing"

Claude Desktop:
[Calls search_skills with query: "test"]
"Found 3 skills:
- test-generator: Generate tests from code
- test-runner: Execute test suites
- api-test: Test API endpoints"

🔒 Security Features

Read Restrictions

Only allows reading from approved directories:
- .claude-skills/
- .ai-factory/
- projects/
Path traversal protection (.. not allowed)

Write Restrictions

Can only write to .ai-factory/ directory
Cannot modify skills or projects
Cannot access system files

Execution Restrictions

Only read-only commands allowed
Whitelist: help, status, list, show, check, get, view, info, summary, report
5 second timeout on all executions

🐛 Troubleshooting

Server Not Starting

# Check if Node.js is installed
node --version  # Should be v16+

# Test the server manually
cd ~/.claude-skills/core/mcp-server.skill
node scripts/start.js
# Should output: "✅ AI Factory MCP Server ready"
# Press Ctrl+C to stop

Claude Desktop Not Seeing Tools

Check config file location:
- macOS: ~/Library/Application Support/Claude/claude_desktop_config.json
- Windows: %APPDATA%\Claude\claude_desktop_config.json

Verify path in config is correct:

# Test the path
node /path/from/config/start.js

Restart Claude Desktop completely
Check logs:
- macOS: ~/Library/Logs/Claude/mcp*.log
- Windows: %APPDATA%\Claude\logs\mcp*.log

Permission Errors

# Make sure script is executable
chmod +x ~/.claude-skills/core/mcp-server.skill/scripts/start.js

# Check ownership
ls -la ~/.claude-skills/core/mcp-server.skill/

Tools Not Working

Check that directories exist:

ls -la ~/.claude-skills/core/
ls -la ~/.ai-factory/
ls -la ~/projects/

Verify skill names are correct:

ls ~/.claude-skills/core/ | grep task

📊 How It Works

MCP Protocol Flow

1. Claude Desktop starts → Launches MCP server process
2. MCP server connects via stdio (standard input/output)
3. Claude Desktop requests: "What tools are available?"
4. MCP server responds: "Here are 16 tools..."
5. User asks: "List my skills"
6. Claude Desktop decides: "I'll use the skills_list tool"
7. Claude calls: tools/call with name="skills_list"
8. MCP server reads ~/.claude-skills/core/
9. MCP server returns: JSON with all skills
10. Claude Desktop shows: Formatted response to user

Why This is Better Than REST API

Feature	MCP Server	REST API
Setup	One config file	Tunnel + URL sharing
Auth	Built-in (stdio)	Token management
Discovery	Automatic	Manual
Integration	Native Claude	External tool
Performance	Direct (no HTTP)	Network latency
Security	Process isolation	Network exposure

🎯 Next Steps

Extend with More Tools

You can easily add more tools by editing scripts/start.js:

// Add to tools/list handler:
{
  name: 'my_custom_tool',
  description: 'Does something awesome',
  inputSchema: {
    type: 'object',
    properties: {
      param1: { type: 'string' }
    },
    required: ['param1']
  }
}

// Add to tools/call handler:
case 'my_custom_tool': {
  const { param1 } = args;
  // Your logic here
  return {
    content: [{ type: 'text', text: 'Result!' }]
  };
}

Package as Product

This MCP server can be packaged and sold:

Bundle with installer
Add licensing
White-label for enterprises
Create marketplace listing

Build More MCP Servers

Create specialized MCP servers:

bmad-mcp - BMAD agent orchestration
skill-factory-mcp - Automated skill generation
analytics-mcp - Deep analytics tools
git-mcp - Git operations via MCP

🌟 The Impact

Before MCP Server:

You ↔ Claude.ai (chat) ↔ Manual copy/paste ↔ Claude Code
Time: Minutes per operation

After MCP Server:

You ↔ Claude Desktop ↔ MCP Server ↔ Your Files
Time: Seconds per operation

Result: 10x faster collaboration! 🚀

📚 References

🎉 You Did It!

You now have a production-ready MCP server that:

✅ Exposes all AI Factory capabilities
✅ Works with Claude Desktop
✅ Follows official MCP protocol
✅ Is secure by design
✅ Can be extended infinitely
✅ Is packageable as a product

Welcome to the future of AI-native development! 🚀

mcp-server

SKILL.md