mcp-chatbot

This MCP Chatbot demonstrates how to compose an interactive CLI assistant that leverages multiple MCP servers as dynamic tool providers. Tools from configured MCP servers are discovered at runtime and exposed to the language model through a system prompt, enabling the LLM to decide when to call a tool and which one to use. The client (Python) handles tool discovery, tool formatting for the LLM, and tool execution, while the LLM interprets responses and decides whether to perform a tool call or return a direct answer. You can interact with the chatbot by starting the client and asking questions that may be augmented by the available tools, such as querying data from a database or running a headless browser task via the Puppeteer service.

To use it, first ensure your environment is configured, then run the client (python main.py). The assistant will automatically detect the tools exposed by the configured servers and incorporate them into its available capabilities. When you ask a question that requires a tool, the LLM will indicate a tool call, the client will execute the tool, and the result will be fed back to the LLM for final presentation to you.

Prerequisites:

• Python 3.10+ (tested with Python 3.10)
• pip (Python package installer)
• git
• Optional: a compatible LLM provider key (e.g., OpenAI-compatible API key)

Installation steps:

1. Clone the repository and install dependencies

   git clone https://github.com/3choff/mcp-chatbot.git
   cd mcp-chatbot
   

2. Create a virtual environment (recommended) and install requirements

   python -m venv venv
   # Windows
   venv\Scripts\activate
   # macOS/Linux
   source venv/bin/activate
   
   pip install -r requirements.txt
   

3. Set up environment variables Create a .env file in the project root with your API key(s). Example:

   LLM_API_KEY=your_api_key_here
   

4. Configure MCP servers Edit or create servers_config.json to define the MCP servers you want to use. Example provided in README:

   {
     "mcpServers": {
       "sqlite": {
         "command": "uvx",
         "args": ["mcp-server-sqlite", "--db-path", "./test.db"]
       },
       "puppeteer": {
         "command": "npx",
         "args": ["-y", "@modelcontextprotocol/server-puppeteer"]
       }
     }
   }
   

   You may add environment variables under each server using the env field as needed.

5. Run the client

   python main.py
   

Tips and considerations:

• The system prompt auto-includes tool descriptions for LLM awareness; ensure your MCP servers expose recognizable tools for best results.
• If a tool call fails, check that the corresponding MCP server is running and reachable, and inspect any environment variables or command-line arguments required by that server.
• You can customize environment variables per server in servers_config.json by adding an env object, e.g., "env": { "API_KEY": "your_api_key_here" }
• The project is Python-based and uses uvicorn-compatible MCP servers; if you switch to a different server implementation, ensure it adheres to MCP conventions so the tool discovery and formatting work correctly.