Smolagents MCPs

🚀 SmolAgents + MCP Server

This project demonstrates the use of the SmolAgents library to build an AI assistant capable of performing web searches and scraping/crawling websites, achieved through the integration of Brave Search and the Crawl4ai Model Context Protocol (MCP) server.

🚀 Quick Start

✨ Features

• Build an AI assistant using the SmolAgents library.
• Perform web searches and scrape/crawl websites.
• Integrate with Brave Search and the Crawl4ai MCP server.
• Modular design for easy extension of agent capabilities.

📦 Installation

Before running this project, ensure you have installed the following dependencies:

• smolagents
• python-dotenv
• mcp
• @modelcontextprotocol/server-brave-search (globally installed via npm/npx)
• litellm

You can install the Python packages using pip:

pip install -r requirements.txt

Make sure you have Node.js, npm/npx, and uv installed to run the MCP server.

💻 Usage Examples

Basic Usage

1. Clone the repository or download the script (smolagent_with_mcp.py).
2. Navigate to the project directory in the terminal.
3. Run the script:

python smolagent_with_mcp.py

4. Interact with the assistant by entering search queries. Type exit or quit to terminate the session.

The code encapsulates the agent's initialization and the main interaction loop in a Python class (SmolAgentWithMCP). This object-oriented approach provides a modular design and makes it easier to extend the agent's capabilities, such as integrating more MCP servers and their tools.

Advanced Usage

The smolagent_with_mcp.py file's structure allows for easily adding more MCP servers and their tools in a modular way. To add a new MCP server:

1. Define the StdioServerParameters for the new server. This involves specifying the command to run the server, its arguments, and any necessary environment variables (similar to how brave_server_params and crawl4ai_server_params are defined).
2. Add the newly defined server parameter object to the server_list in the main function. The BraveSearchAgentApp class will automatically initialize the new server and include its tools in the agent's capabilities.

Example:

# Define parameters for the new server (replace with actual commands and arguments)
new_server_params = StdioServerParameters(
    command="your_server_command",
    args=["arg1", "arg2"],
    env={"YOUR_API_KEY": os.getenv("YOUR_API_KEY")},
)

def main():
    # Define the list of MCP server parameters
    server_list = [
        brave_server_params,
        crawl4ai_server_params,
        new_server_params
    ]
    
    # Initialize the agent and run
    agent = SmolAgentWithMCP(server_list)
    agent.run()

if __name__ == "__main__":
    main()

📚 Documentation

Environment Variables

Create a .env file in the project's root directory and add the following variables:

BRAVE_API_KEY=your_brave_api_key
OPENAI_API_KEY=your_openai_api_key

Replace your_brave_api_key and your_openai_api_key with your actual API keys.

Notes

• Ensure that npx is installed and available in the system's PATH, as it is required to run the Brave Search MCP server.
• The ToolCallingAgent in SmolAgents is designed to output tool calls in a structured format (e.g., JSON), which are then executed by the environment (in this case, the main loop in the main script uses MCPClient for execution).

🔧 Technical Details

The project uses the SmolAgents library to build an AI assistant. It integrates with the Brave Search and Crawl4ai MCP servers to enable web search and scraping/crawling capabilities. The agent's initialization and interaction loop are encapsulated in a Python class, providing a modular design. The StdioServerParameters class is used to define the parameters for running MCP servers, and the BraveSearchAgentApp class initializes these servers and includes their tools in the agent's capabilities.

📄 License

This project is licensed under the MIT License, which allows you to freely use, modify, and distribute the code.