File size: 2,463 Bytes
b58d21f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 |
# 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:
```bash
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:
```bash
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:
```python
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:
```python
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:
```python
async def shutdown():
mcp_client.disconnect()
```
## Customization
You can modify the `CodeAgent` configuration by changing the model or MCP server parameters. For example:
```python
# 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(),
)
```
|