a2a-mcp

A2A-MCP: Agent-to-Agent Model Control Protocol

A sophisticated multi-agent system that combines the Model Context Protocol (MCP) with Agent-to-Agent (A2A) communication for intelligent text processing, web interaction, and automated workflows.

🌟 Features

🤖 Intelligent Multi-Agent System

• Smart Agent Coordination: Automatic detection of user intent and routing to appropriate agents
• A2A Communication: Direct agent-to-agent communication for complex workflows
• MCP Integration: Standards-compliant Model Context Protocol implementation

🔧 Core Services (MCP Tools)

• Web Search & Weather: DuckDuckGo integration with weather-specific queries
• Website Content Extraction: Headless browser-based text extraction
• Time & Date Services: NTP-synchronized accurate time information
• File Processing: Multi-format document conversion to PDF
• Data Anonymization: Intelligent PII detection and removal

🎯 Specialized Agents (A2A)

• Text Optimizer: Professional email generation and tone adjustment
• Grammar Corrector (Lektor): German/English grammar and spelling correction
• Sentiment Analysis: Emotion detection and sentiment scoring
• Query Refactoring: LLM-optimized query reformulation
• User Interface Agent: Intelligent request interpretation and routing

🌐 User Interfaces

• Gradio Web Interface: User-friendly browser-based interaction
• RESTful API: Programmatic access to all services
• CLI Integration: Command-line tool compatibility

🏗️ Architecture

┌─────────────────────────────────────────────────────────────┐
│                    A2A-MCP System                          │
├─────────────────────────────────────────────────────────────┤
│  Gradio Interface (Port 7860)                              │
│  ├── File Upload & Processing                              │
│  ├── Natural Language Input                                │
│  └── Tonality Selection                                    │
├─────────────────────────────────────────────────────────────┤
│  User Interface Agent (Intelligent Router)                 │
│  ├── Intent Detection                                      │
│  ├── Agent Selection                                       │
│  └── Response Coordination                                 │
├─────────────────────────────────────────────────────────────┤
│  MCP Server (Port 8000)                     A2A Registry   │
│  ├── Web Search & Weather                   ├── Optimizer  │
│  ├── Website Extraction                     ├── Lektor     │
│  ├── Time/Date Services                     ├── Sentiment  │
│  ├── File Conversion                        ├── Query Ref  │
│  └── Anonymization                          └── UI Agent   │
├─────────────────────────────────────────────────────────────┤
│  Backend Services                                          │
│  ├── Selenium WebDriver                                    │
│  ├── DuckDuckGo Search                                     │
│  ├── NTP Time Sync                                         │
│  ├── PDF Conversion                                        │
│  └── LLM Integration (Ollama)                              │
└─────────────────────────────────────────────────────────────┘

🚀 Quick Start

Prerequisites

• Python 3.11+
• Ollama with qwen2.5:latest model
• Modern web browser (for Gradio interface)

Installation

1. Clone the repository:

   git clone <repository-url>
   cd a2a_mcp
   

2. Install dependencies:

   # Using uv (recommended)
   uv sync
   
   # Or using pip
   pip install -r requirements.txt
   

3. Configure environment:

   cp .env.example .env
   # Edit .env with your settings
   

4. Start Ollama (if not running):

   ollama serve
   ollama pull qwen2.5:latest
   

Quick Launch

Start all services with the integrated launcher:

python launcher.py

This will automatically start:

• MCP Server on http://localhost:8000
• Gradio Interface on http://localhost:7860
• A2A Agent Registry (embedded)

🎯 Usage Examples

Web Interface

1. Open your browser to http://localhost:7860
2. Try these example requests:

"Wie wird das Wetter morgen in Berlin?"
→ Automatic web search with weather optimization

"Korrigiere diesen Text: Das ist ein sehr schlechte Satz mit viele Fehler."
→ Grammar correction via Lektor agent

"Optimiere diesen Text für eine professionelle E-Mail: Das Produkt ist Schrott!"
→ Professional email generation via Optimizer agent

"Analysiere das Sentiment: Ich bin so glücklich über dieses großartige Produkt!"
→ Sentiment analysis with emotion detection

"Wie spät ist es jetzt?"
→ NTP-synchronized time retrieval

API Usage

import httpx

# Direct MCP tool call
async with httpx.AsyncClient() as client:
    response = await client.post(
        "http://localhost:8000/mcp/call-tool",
        json={
            "name": "duckduckgo_search",
            "arguments": {"query": "weather Berlin", "max_results": 5}
        }
    )
    print(response.json())

Agent Integration

from agent_server.user_interface import process_input

# Intelligent request processing
result = await process_input("Mache diesen Text freundlicher: Ihre Anfrage wurde abgelehnt.")
print(result.final_result)

🔧 Configuration

Environment Variables

# LLM Configuration
BASE_URL=http://localhost:11434/v1
API_KEY=ollama
USER_INTERFACE_MODEL=qwen2.5:latest
OPTIMIZER_MODEL=qwen2.5:latest
LEKTOR_MODEL=qwen2.5:latest
SENTIMENT_MODEL=qwen2.5:latest
QUERY_REF_MODEL=qwen2.5:latest

# Server Configuration
SERVER_HOST=localhost
SERVER_PORT=8000
SERVER_SCHEME=http
GRADIO_HOST=127.0.0.1
GRADIO_PORT=7860

# Debug Options
DEBUG_AGENT_RESPONSES=false
DEBUG_A2A_CALLS=false

# Service Configuration
ANONYMIZER_USE_LLM=false
ANONYMIZER_LLM_ENDPOINT=
ANONYMIZER_LLM_API_KEY=
ANONYMIZER_LLM_MODEL=

Model Requirements

Ensure these models are available in Ollama:

ollama pull qwen2.5:latest  # Primary model for all agents
# Or configure different models per agent in .env

📚 Available Tools & Agents

MCP Tools

Tool	Description	Parameters
get_current_time	NTP-synchronized UTC time	None
duckduckgo_search	Web search with weather optimization	query, max_results
extract_website_text	Extract main content from URLs	url
anonymize_text	Remove PII from text	text
convert_to_pdf	Convert files to PDF format	input_filepath, output_directory

A2A Agents

Agent	Purpose	Input	Output
Optimizer	Professional text optimization	Raw text + tonality	Polished professional text
Lektor	Grammar & spelling correction	Text with errors	Corrected text
Sentiment	Emotion & sentiment analysis	Any text	Sentiment score + emotions
Query Ref	LLM query optimization	User query	Optimized query
User Interface	Intelligent request routing	Natural language	Coordinated response

Supported File Types

Text Processing

• .txt, .md, .py, .csv, .log, .json, .xml, .html

PDF Conversion

• Text files: All above formats
• Images: .jpg, .png, .gif, .bmp, .tiff, .webp
• Office docs: .docx, .xlsx, .pptx (requires LibreOffice)

🔄 Workflows

Intelligent Text Processing

User Input → Intent Detection → Agent Selection → Processing → Response
     ↓              ↓              ↓              ↓          ↓
"Fix grammar" → Text Processing → Lektor Agent → Correction → Clean Text

Professional Email Generation

Complaint Text → Optimizer Agent → Professional Email → Lektor Check → Final Email

Multi-Step Analysis

Raw Text → Query Refactor → Optimization → Grammar Check → Sentiment Analysis

🛠️ Development

Project Structure

a2a_mcp/
├── agent_server/           # A2A agents
│   ├── user_interface.py   # Main coordination agent
│   ├── optimizer.py        # Text optimization
│   ├── lektor.py          # Grammar correction
│   ├── sentiment.py       # Sentiment analysis
│   └── query_ref.py       # Query refactoring
├── mcp_services/          # MCP service implementations
│   ├── mcp_search/        # DuckDuckGo integration
│   ├── mcp_website/       # Web scraping
│   ├── mcp_time/         # NTP time services
│   ├── mcp_anonymizer/   # Data anonymization
│   └── mcp_fileconverter/ # PDF conversion
├── mcp_main.py           # MCP server
├── launcher.py           # Service orchestrator
├── gradio_interface.py   # Web UI
├── a2a_server.py        # A2A registry
└── uploaded_files/      # File upload storage

Adding New Agents

1. Create agent file in agent_server/:

from pydantic_ai import Agent
from pydantic import BaseModel

class MyAgentResponse(BaseModel):
    result: str

async def my_agent_a2a_function(messages: list) -> MyAgentResponse:
    # Implementation
    pass

2. Register in A2A server:

# In a2a_server.py
registry.register_a2a_agent("my_agent", my_agent_a2a_function)

3. Add to user interface agent routing logic

Adding MCP Tools

1. Implement service in mcp_services/
2. Add endpoint in mcp_main.py
3. Register tool in MCP configuration

🧪 Testing

Manual Testing

# Test individual agents
python agent_server/sentiment.py
python agent_server/optimizer.py

# Test MCP server
curl http://localhost:8000/health

# Test full workflow
python a2a_server.py

Example Test Cases

# Sentiment analysis
await sentiment_agent("I love this amazing product!")

# Text optimization
await optimizer_agent("Das Produkt ist Schrott!", tonality="professionell")

# Grammar correction
await lektor_agent("Das ist ein sehr schlechte Satz.")

🤝 Contributing

1. Fork the repository
2. Create a feature branch: git checkout -b feature/amazing-feature
3. Commit changes: git commit -m 'Add amazing feature'
4. Push to branch: git push origin feature/amazing-feature
5. Open a Pull Request

📄 License

This project is licensed under the AGPL v3 License - see the License.md file for details.

🙏 Acknowledgments

• Pydantic AI for the agent framework
• Model Context Protocol for the standards
• Gradio for the web interface
• Ollama for local LLM support

---

Built with ❤️ for intelligent multi-agent workflows