Content
# nano-MCP: Model Context Protocol Implementation
An implementation of the Model Context Protocol (MCP) for tool-augmented Language Models, enabling LLMs like GPT-4.1 and Llama models to use external tools to complete tasks autonomously.
## Architecture
This project follows the Model Context Protocol (MCP) architectural pattern:
- **LLM Integration**: Logic for interacting with LLM APIs (OpenAI, Claude, etc.). Sends prompts, parses tool call outputs, and routes tool calls to MCP client.
- **MCP Client**: Routes tool calls to the appropriate MCP servers.
- **MCP Servers**: Dockerized FastAPI applications exposing tool APIs:
- File Management Server (read, write, tree operations)
- Version Control Server (git operations, command execution)
## Components
- `src/llm/llm_mcp_client.py`: The core client for LLM integration with MCP tools
- `src/llm/host_cli.py`: Command-line interface for interacting with the LLM+MCP system
- `src/llm/host.py`: FastAPI server providing a web interface at `/`
## Usage
## Docker Setup to run MCP Servers
Create a `shared_data` folder in the `servers` folder.
The project uses Docker Compose to orchestrate the MCP servers:
```bash
cd src/servers/
docker-compose up -d
```
This will start:
- File Management Server on port 8000
- Version Control Server on port 8001
## MCP Client Hub
### Install Requirements
```
# activate your environment
cd src/
pip install -r requirements.txt
```
### Start the MCP Client Hub
```bash
cd src/client
export MCP_SERVER_URLS="http://localhost:8002,http://localhost:8003"
python nano_mcp_client.py
```
## MCP CLI or UI
### Environment Variables
Configure the following environment variables:
```
OPENAI_API_KEY=your_api_key
OPENAI_BASE_URL=https://api.openai.com/v1
MODEL_NAME=gpt-4-turbo
HOST_MODEL=openai
MCP_CLIENT_URL=http://localhost:8001
```
### CLI Interface
Run the CLI interface:
```bash
cd src/llm
python host_cli.py
```
You can also provide configuration via command-line arguments:
```bash
python host_cli.py --api-key YOUR_API_KEY --model gpt-4-turbo --mcp-url http://localhost:8001
```
### Web Interface
Start the web server:
```bash
cd src/llm
python host.py
```
The web interface will be available at http://localhost:7899/
## Protocol Details
The Model Context Protocol (MCP) allows LLMs to:
1. Discover available tools from MCP servers
2. Call tools with proper arguments
3. Receive and process tool results
4. Continue execution in an autonomous loop
Each MCP server exposes two main endpoints:
- `GET /list/tools`: Lists available tools with schemas
- `POST /execute/call`: Executes tools based on name and input
## Development
To add new tools, create a new server in the `servers/` directory implementing the MCP protocol endpoints.
MCP Store
