Content
# A2A-Using-MCP
This project demonstrates a multi-agent system using Google's Gemini LLM and the Model Contex Protocol (MCP) to automate lead creation and verification. It features specialized Sales, Finance, and Legal agents.
## Overview
The system automates a lead processing workflow:
1. A **Sales Agent** creates a new lead in a CRM (e.g., Salesforce) based on user input.
2. This triggers an automated verification process:
* The **Finance Agent** assesses financial viability.
* If approved, the **Legal Agent** checks for legal/compliance acceptability.
3. The Sales Agent updates the lead's status in the CRM based on these verifications and informs the user.
The Sales Agent's client-side logic orchestrates this Agent-to-Agent (A2A) communication.
## Features
* **Specialized AI Agents:** Dedicated agents for Sales (CRM), Finance (financial assessment), and Legal (legal assessment).
* **LLM-Powered Tools:** Agents use Gemini-powered tools for their tasks.
* **MCP-Based Communication:** Agents interact via the Model Contex Protocol.
* **Orchestrated Workflow:** The Sales Agent's client automatically manages the multi-step verification process after lead creation.
* **Salesforce Integration:** Configurable integration with Salesforce for CRM operations.
## Core Components
* **`salesforce_mcp_server.py`**: Sales Agent's server (Salesforce interaction).
* **`finance_mcp_server.py`**: Finance Agent's server (financial assessment).
* **`legal_mcp_server.py`**: Legal Agent's server (legal assessment).
* **`client_gemini.py`**: Sales Agent's client & A2A orchestrator for lead verification.
* **`.env`**: For environment variables (API keys, credentials).
* **`requirements.txt`**: Python package dependencies.
## Setup and Installation
### Prerequisites
* Python 3.9+
* Git
* Google Gemini API Key
* Salesforce Developer Org/Sandbox (optional, for full functionality)
### Steps
1. **Clone Repository:**
```bash
git clone https://github.com/Krishnapopat/A2A-Using-MCP.git
cd A2A-Using-MCP
```
2. **Virtual Environment (Recommended):**
```bash
python -m venv venv
# Windows: venv\Scripts\activate
# macOS/Linux: source venv/bin/activate
```
3. **Install Dependencies:**
```bash
pip install -r requirements.txt
```
4. **Environment Variables (`.env` file):**
```env
# For Google Gemini API (used by all agents)
GOOGLE_API_KEY="YOUR_GOOGLE_API_KEY"
# For Salesforce Integration (used by salesforce_mcp_server.py)
SALESFORCE_USERNAME="your_salesforce_username"
SALESFORCE_PASSWORD="your_salesforce_password"
SALESFORCE_SECURITY_TOKEN="your_salesforce_security_token"
```
## Running the System
The Sales Agent client (`client_gemini.py`) orchestrates the process. It will attempt to start the Finance and Legal agent MCP server scripts (`finance_mcp_server.py`, `legal_mcp_server.py`) as subprocesses when needed.
1. **Ensure Server Scripts Are Present:**
Verify `salesforce_mcp_server.py`, `finance_mcp_server.py`, and `legal_mcp_server.py` are in the same directory as `client_gemini.py` (or update paths in `client_gemini.py`).
2. **Run the Sales Agent Client:**
```bash
python client_gemini.py salesforce_mcp_server.py
```
*(The `salesforce_mcp_server.py` argument can be omitted if it's the default in `client_gemini.py`'s `main` function).*
The client will start, connect to (and start) `salesforce_mcp_server.py`. When you request a lead creation, the client will then dynamically start and interact with the Finance and Legal agent servers.
## Workflow Example (Simplified)
1. **User (to Sales Agent):** "Create a lead for Innovate Corp..."
2. **Sales LLM:** Decides to use `create_record`.
3. **`client_gemini.py` (Orchestrator):**
* Executes `create_record` via `salesforce_mcp_server.py`.
* If successful, starts `finance_mcp_server.py` & calls its assessment tool.
* If Finance approves, starts `legal_mcp_server.py` & calls its assessment tool.
* Calls `salesforce_mcp_server.py` again to update the lead status (e.g., to "Fully Verified").
* Compiles a summary.
4. **Sales LLM:** Receives the summary.
5. **Sales LLM (to User):** "OK. Lead for Innovate Corp created and verified (ID 00Q...). Status: Fully Verified."
## Contributing
Contributions are welcome! Please fork, make changes, and submit pull requests. Open an issue for bugs or suggestions.
## License
This project is licensed under the [MIT License](LICENSE).
