# MCP Agent Selection Service

Intelligent agent selection service for Claude Code. Provides context-aware agent recommendations and automated project bootstrapping with specialist teams.

## Installation & Setup

### 1. Install Dependencies
```bash
uv sync
```

### 2. Test the Server
```bash
# Test agent library functionality
python test_agents.py

# Test production server
uv run server
```

### 3. Add to Claude Code for Local Development

Add the MCP server to your Claude Code configuration:

```bash
claude mcp add agent-selection -- uv run server
```

This will automatically configure the server with the correct working directory. You can also manually edit your `~/.claude/settings.json`:

```json
{
  "mcpServers": {
    "agent-selection": {
      "command": "uv",
      "args": ["run", "server"],
      "cwd": "/home/rpm/claude/mcp-agent-selection"
    }
  }
}
```

## Available MCP Tools

### Core Tools
- `set_project_roots` - Focus on specific directories
- `get_current_roots` - Check current project focus  
- `clear_project_roots` - Remove project focus
- `recommend_agents` - Get intelligent agent recommendations
- `get_agent_content` - Retrieve full agent templates
- `list_agents` - Browse agent catalog
- `server_stats` - Get server statistics

### Hierarchy Navigation (NEW)
- `get_sub_agents` - List specialists for composed agents
- `get_agent_hierarchy` - Show complete parent-child structure
- `get_parent_agent` - Get parent context for sub-agents

## Usage Examples

### Basic Recommendations
```bash
# Through Claude Code MCP interface
recommend_agents({
  "task": "I need help with Python FastAPI development",
  "limit": 3
})
```

### Project-Focused Development
```bash
# Set project roots for targeted suggestions
set_project_roots({
  "directories": ["src/api", "src/backend"],
  "base_path": "/path/to/project",
  "description": "API development focus"
})

# Get recommendations - now context-aware
recommend_agents({
  "task": "optimize database queries",
  "project_context": "FastAPI backend"
})
```

### Browse Available Specialists
```bash
# List all agents
list_agents()

# Search for specific expertise  
list_agents({"search": "testing"})
```

## Features

- **36+ Specialist Agents** loaded from templates
- **Composed Agent Architecture** - hierarchical specialists with flat access
- **Intelligent Recommendations** with confidence scoring  
- **Project Roots Support** for focused analysis
- **Context-Aware Suggestions** based on your work
- **Hierarchical Navigation** - discover related specialists
- **Real-time Agent Library** with automatic updates

## Agent Library

Serves agents from local `agent_templates/` directory including:

- 🎭-subagent-expert
- 🔮-python-mcp-expert  
- 🚄-fastapi-expert
- 🐳-docker-infrastructure-expert
- 🧪-testing-integration-expert (composed agent with 2 specialists)
- 📖-readme-expert
- And 30+ more specialists...

Perfect for bootstrapping new projects with the right expert team!