SmolAgents + MCP servers

# SmolAgents + MCP servers This project demonstrates building an AI assistant using the SmolAgents library that can perform web searches and scrape/crawl websites by integrating with the Brave Search and the Crawl4ai Model Context Protocols (MCP) servers. ## Requirements Before running this project, ensure you have the following dependencies installed: - `smolagents` - `python-dotenv` - `mcp` - `@modelcontextprotocol/server-brave-search` (installed globally via npm/npx) - `litellm` You can install the Python packages using pip: ```bash pip install -r requirements.txt ``` Ensure you have Node.js and npm/npx and uv installed to run the MCP servers. ## Environment Variables Create a `.env` file in the root directory of the project 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. ## Usage 1. Clone the repository or download the script (`smolagent_with_mcp.py`). 2. Navigate to the project directory in your terminal. 3. Run the script: ```bash python smolagent_with_mcp.py ``` 4. Interact with the assistant by typing your search queries. Type `exit` or `quit` to terminate the session. the code encapsulates the agent initialization and the main interaction loop within a Python class (`SmolAgentWithMCP`). This object-oriented approach offers modularity and makes it easier to extend the agent's capabilities, such as integrating additional MCP servers and their tools. ## Notes * Ensure that `npx` is installed and available in your 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 (like JSON) which are then executed by the environment (in this case, the script's main loop using the `MCPClient`). ## Adding More MCP Servers The `smolagent_with_mcp.py` file is structured to allow 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 parameters 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: ```python # Define parameters for a new server (replace with actual command and args) 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 # Add the new server parameters here # Add other server parameters here as needed ] app = BraveSearchAgentApp(server_list) app.run() if __name__ == "__main__": main() ``` ## MCP used in the project: https://playbooks.com/mcp/ritvij14-crawl4ai https://github.com/modelcontextprotocol/servers/tree/main/src/brave-search ## License This project is licensed under the MIT License. See the `LICENSE` file for more details.