examples/smolagents_benchmark/README.md

Smolagents Chat Server Demo

This is a simple web server that provides a chat interface for interacting with an AI code agent powered by smolagents and the Qwen2.5-Coder-32B-Instruct model, enhanced with MCP (Model Control Protocol) tools.

Features

• Web-based chat interface
• AI code agent powered by Qwen2.5-Coder
• Integration with MCP tools through MCPClient
• Asynchronous request handling
• Clean, responsive UI
• Graceful shutdown handling

Requirements

• Python 3.8+
• Starlette
• AnyIO
• Smolagents with MCP support

Installation

1. Install the required packages:

pip install starlette anyio smolagents[mcp] uvicorn

2. Optional: If you want to use a specific model, you may need additional dependencies.

Usage

1. Run the server:

uvicorn examples.server.main:app --reload

2. Open your browser and navigate to http://localhost:8000

3. Interact with the AI code agent through the chat interface

How It Works

The server consists of two main routes:

• / - Serves the HTML page with the chat interface
• /chat - API endpoint that processes messages and returns responses

The server integrates with MCP tools through the following components:

1. MCPClient Configuration:

mcp_server_parameters = {
    "url": "https://evalstate-hf-mcp-server.hf.space/mcp",
    "transport": "streamable-http",
}
mcp_client = MCPClient(server_parameters=mcp_server_parameters)

2. CodeAgent with MCP Tools:

agent = CodeAgent(
    model=InferenceClientModel(model_id="Qwen/Qwen2.5-Coder-32B-Instruct"),
    tools=mcp_client.get_tools(),
)

When a user sends a message:

1. The message is sent to the /chat endpoint
2. The server runs the AI code agent in a separate thread
3. The agent processes the message using MCP tools
4. The agent's response is returned to the client and displayed in the chat

The server also includes a shutdown handler that properly disconnects the MCP client when the server stops:

async def shutdown():
    mcp_client.disconnect()

Customization

You can modify the CodeAgent configuration by changing the model or MCP server parameters. For example:

# Custom MCP server
mcp_server_parameters = {
    "url": "your-mcp-server-url",
    "transport": "your-transport-method",
}

# Custom agent configuration
agent = CodeAgent(
    model=InferenceClientModel(model_id="your-preferred-model"),
    tools=mcp_client.get_tools(),
)