Model Context Protocol (mcp) Server

🚀 MCP Client with LangChain / Python

This simple Model Context Protocol (MCP) client demonstrates the use of MCP server tools by a LangChain ReAct Agent. It utilizes the utility function convert_mcp_to_langchain_tools() from the langchain_mcp_tools package, which can initialize multiple specified MCP servers in parallel and convert their available tools into a LangChain-compatible tool list (List[BaseTool]).

Currently, it supports large language models (LLMs) from Anthropic, OpenAI, and Groq.

The TypeScript version of the MCP client can be found here: here

🚀 Quick Start

Prerequisites

• Python 3.11+
• [Optional] Install uv (or uvx) to run the MCP server based on Python packages.
• [Optional] Install npm 7+ (or use npx) to run the MCP server based on Node.js packages.
• Obtain API keys from Anthropic, OpenAI, and/or Groq as needed.

📦 Installation

1. Install dependencies:

make install

2. Set API keys:

cp .env.template .env

• Update the .env file as needed.
• .gitignore is configured to ignore .env to prevent accidental credential commits.

3. Configure the LLM and MCP server settings in llm_mcp_config.json5 if necessary.

• Configuration file format. For MCP servers, follow the same structure as Claude for Desktop, with the only difference being changing the key name mcpServers to mcp_servers to conform to the snake_case naming convention commonly used in JSON configuration files.
• The file format is JSON5, which allows comments and trailing commas.
• The format is further extended to replace corresponding environment variable values with ${...} placeholders.
• Keep all credentials and private information in the .env file and reference the required values using the ${...} notation.

💻 Usage Examples

Basic Usage

Run the application:

make start

The first run may take some time.

Advanced Usage

Run in debug mode:

make start-v

View command-line options:

make start-h

In the prompt, you can directly press Enter to use example queries, which will execute MCP server tool calls. Example queries can be configured in llm_mcp_config.json5.