Agent Factory

A universal Model Context Protocol (MCP) server that automatically discovers and registers CLI tools as MCP tools. Transform any CLI tool (Qwen, Ollama, Aider, etc.) into an AI-powered agent with persona configuration.

Features

• Auto-Discovery: Automatically parse CLI --help output to generate tool metadata
• Zero-Code Registration: Register tools via config file or command-line arguments
• Persona Support: Configure system prompts to create specialized AI agents
• Multi-Provider: Use multiple AI tools simultaneously (Qwen, Gemini, Aider, etc.)
• Runtime Registration: Add new tools dynamically via MCP protocol

Architecture

graph TB
    subgraph "MCP Client"
        A[Claude Desktop / Claude Code]
    end

    subgraph "Agent Factory MCP Server"
        B[Server Entry Point]
        C[Config Loader]
        D[Tool Registry]
        E[Dynamic Tool Factory]

        subgraph "Providers"
            F[QwenProvider]
            G[GenericCliProvider]
        end

        subgraph "Parsers"
            H[HelpParser]
        end
    end

    subgraph "CLI Tools"
        I[qwen]
        J[gemini]
        K[aider]
        L[ollama]
        M[...any CLI tool]
    end

    A -->|stdio| B
    B --> C
    B -->|CLI args| G
    C -->|load config| D
    G -->|create| D
    D --> E
    E -->|generate| F
    F -->|execute| I
    F -->|execute| J
    F -->|execute| K
    G -->|parse --help| H
    H -->|metadata| G

State Transition

stateDiagram-v2
    [*] --> Initialization

    Initialization --> LoadConfig: Start
    Initialization --> ProcessCLIArgs: CLI args provided

    LoadConfig --> ProcessCLIArgs: Config loaded
    ProcessCLIArgs --> RegisterProviders

    RegisterProviders --> ProviderCreated: Tool available
    RegisterProviders --> ProviderSkipped: Tool not found

    ProviderCreated --> GenerateTools
    ProviderSkipped --> RegisterProviders: Next tool

    GenerateTools --> ToolRegistered
    ToolRegistered --> RegisterProviders: Next tool

    RegisterProviders --> ServerRunning: All tools processed
    ServerRunning --> [*]: Ready for MCP requests

    ServerRunning --> RuntimeRegistration: register_cli_tool called
    RuntimeRegistration --> ServerRunning: Tool added

    note right of LoadConfig
        Loads ai-tools.json
        or .qwencoderc.json
    end note

    note right of ProcessCLIArgs
        Parses CLI args like:
        npx agent-factory-mcp qwen gemini aider
    end note

Installation

# Install via npm
npm install -g agent-factory-mcp

# Or use with npx without installation
npx agent-factory-mcp

# Or use with bun
bunx agent-factory-mcp

Configuration

Method 1: Command-Line Arguments

Register tools directly via CLI arguments:

npx agent-factory-mcp qwen gemini aider

Method 2: Configuration File

Create ai-tools.json in your project root:

{
  "$schema": "./schema.json",
  "version": "1.0",
  "tools": [
    {
      "command": "qwen",
      "alias": "code-reviewer",
      "description": "Code review expert focusing on security and performance",
      "systemPrompt": "You are a senior code reviewer. Focus on security vulnerabilities, performance issues, and maintainability."
    },
    {
      "command": "qwen",
      "alias": "doc-writer",
      "description": "Technical documentation specialist",
      "systemPrompt": "You write clear, concise technical documentation for developers."
    }
  ]
}

Method 3: Runtime Registration

Use the register_cli_tool MCP tool:

register_cli_tool({
  command: "ollama",
  alias: "local-llm",
  description: "Run local LLM models via Ollama",
  systemPrompt: "You are a helpful AI assistant running locally.",
  persist: true
})

MCP Client Setup

Claude Desktop

Add to your Claude Desktop config:

macOS: ~/Library/Application Support/Claude/claude_desktop_config.json Windows: %APPDATA%\Claude\claude_desktop_config.json Linux: ~/.config/claude/claude_desktop_config.json

{
  "mcpServers": {
    "agent-factory": {
      "command": "npx",
      "args": ["agent-factory-mcp", "qwen", "gemini", "aider"]
    }
  }
}

Claude Code CLI

claude mcp add agent-factory -- npx agent-factory-mcp qwen gemini aider

Usage Examples

Using Specialized Agents

# Code review with security focus
"Use code-reviewer to analyze this file for security issues"

# Documentation generation
"Ask doc-writer to generate API docs for this module"

# General AI assistance
"Use ask-qwen to explain this code"

Multiple AI Tools

# Use different AIs for different tasks
"Use gemini-vision to analyze this screenshot"
"Use aider to refactor this function"
"Use qwen to review the changes"

Configuration Schema

See schema.json for the full configuration schema:

Field	Type	Required	Description
command	string	✅	CLI command to register (e.g., "qwen", "ollama")
enabled	boolean	❌	Whether the tool is enabled (default: true)
alias	string	❌	Custom tool name (default: "ask-{command}")
description	string	❌	Custom tool description
systemPrompt	string	❌	System prompt for AI persona
providerType	string	❌	Provider type: "cli-auto" or "custom"
defaultArgs	object	❌	Default argument values

Development

# Install dependencies
bun install

# Build
bun run build

# Run tests
bun test

# Type check
bun run type-check

# Lint
bun run lint

# Format
bun run format

Project Structure

agent-factory-mcp/
├── src/
│   ├── index.ts              # Server entry point
│   ├── constants.ts          # Constants
│   ├── providers/            # Provider implementations
│   │   ├── base-cli.provider.ts
│   │   ├── generic-cli.provider.ts
│   │   └── qwen.provider.ts
│   ├── tools/                # Tool registry and factory
│   │   ├── registry.ts
│   │   ├── dynamic-tool-factory.ts
│   │   └── simple-tools.ts
│   ├── parsers/              # CLI help parser
│   │   └── help-parser.ts
│   ├── types/                # TypeScript types
│   │   └── cli-metadata.ts
│   └── utils/                # Utilities
│       ├── configLoader.ts
│       ├── commandExecutor.ts
│       ├── logger.ts
│       └── progressManager.ts
├── test/                     # Test files
├── ai-tools.json.example     # Example configuration
├── schema.json               # JSON schema
└── Taskfile.yml              # Task runner configuration

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

MIT License - see LICENSE for details.