MCP CLI LangChain Demo

This project demonstrates how to use the langchain-mcp-adapters library to connect a LangChain agent with multiple Model Context Protocol (MCP) servers, allowing the agent to leverage tools provided by these servers.

Features

• Loads MCP server configurations from a YAML file (mcp_servers_config.yaml).
• Connects to multiple MCP servers concurrently (examples include local Python scripts for Math and Weather, and the official SQLite reference server).
• Integrates MCP tools seamlessly into a LangChain agent using langchain-mcp-adapters.
• Includes a simple demonstration script (main.py) to showcase loading configurations and fetching tools.
• Provides an interactive chat interface (chat_interface.py) powered by a local Ollama model (llama3.3 by default) that can utilize the configured MCP tools.
• The chat interface includes helpful custom commands:
  • /list_servers: List the names of servers defined in the configuration file.
  • /list_tools: List the names of all available tools and the server providing them.
  • /list_tools_details: List detailed information (name, description) for all tools, grouped by server.

Setup

1. Clone this repository:

   git clone <your-repository-url>
   cd mcp_cli_langchain
   

2. Python Environment: Requires Python 3.x. Create and activate a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows use `venv\Scripts\activate`
   

3. Install Dependencies: Install required Python packages:

   pip install -r requirements.txt
   

   (Note: requirements.txt includes python-dotenv which is needed to load the .env file.)
4. Install langchain-mcp-adapters (Editable): This project uses a potentially modified local version of the adapter library. Install it in editable mode:

   # Ensure you are in the project root directory (mcp_cli_langchain)
   cd langchain-mcp-adapters
   pip install -e .
   cd ..
   

5. Clone MCP Reference Servers: The configuration uses the official SQLite reference server. Clone the repository into the project root:

   git clone https://github.com/modelcontextprotocol/servers.git
   

6. Install uv: The SQLite server configuration uses uv to run. Install uv by following the instructions on https://github.com/astral-sh/uv.
7. Setup Ollama:
   • Ensure Ollama is installed and the service is running.
   • Pull the required model (defaults to llama3.3 in chat_interface.py):

     ollama pull llama3.3
     

Configuration

• MCP Servers: MCP server connections are defined in mcp_servers_config.yaml.
  • You can add, remove, or modify server entries in this file.
  • Pay attention to paths (e.g., for local script servers or the cloned servers directory), ensuring they are correct relative to the project root (mcp_cli_langchain).
  • The default configuration includes math, weather, and sqlite servers.
• Ollama: Ollama settings for the chat interface are configured via the .env file in the project root.
  • OLLAMA_MODEL: Specifies the Ollama model to use (e.g., llama3.3).
  • OLLAMA_BASE_URL: The base URL for your running Ollama instance (e.g., http://localhost:11434).
  • OLLAMA_TEMPERATURE: Controls the creativity/randomness of the model's output (e.g., 0.8).
  • Create a .env file if it doesn't exist, based on the example:

  # .env example
  OLLAMA_MODEL=llama3.3
  OLLAMA_BASE_URL=http://localhost:11434
  OLLAMA_TEMPERATURE=0.8
  

Running the Project

Ensure your virtual environment is activated