Langchain MCP Client

🚀 LangChain MCP Client Streaming Application

This Streamlit application provides a user interface to connect to an MCP (Model Context Protocol) server and interact with different LLM providers (OpenAI, Anthropic, Google...).

✨ Features

• Connect to the MCP server via SSE (Server-Sent Events).
• Support single-server and multi-server configurations.
• Choose between different LLM providers such as OpenAI and Claude.
• View, test, and use available MCP tools directly from the UI.
• Have a chat interface to interact with LLM agents.
• Display the results of tool execution.

📦 Installation

1. Clone this repository:

git clone https://github.com/guinacio/langchain-mcp-client.git
cd langchain-mcp-client

2. Create a virtual environment and install dependencies:

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

💻 Usage Examples

Basic Usage

Run the Streamlit application using the following command:

streamlit run app.py

The application will be available at http://localhost:8501.

Advanced Usage

Setting up the MCP Server

To use this application, you need a running MCP server or a valid MCP server URL. For quick testing, you can use the simple MCP server provided in weather_server.py:

1. Install the MCP library:

pip install mcp

2. Run the server:

python weather_server.py

The server will start on the default port 8000. In the Streamlit application, you can connect to it using the URL http://localhost:8000/sse.

📚 Documentation

Configuration Options

LLM Providers

• OpenAI: Requires an OpenAI API key and supports models like gpt-4, gpt-4o, and gpt-3.5-turbo.
• Anthropic: Requires an Anthropic API key and supports the Claude model.
• Google: Requires the Google Generative Language / Vertex AI API.
• Local LLM: Supports local LLMs using Ollama.

MCP Server Connection

• Currently supports SSE (Server-Sent Events) connections.
• Enter the SSE endpoint URL of your MCP server (e.g., http://localhost:8000/sse).

Server Modes

• Single Server: Connect to one MCP server.
• Multiple Servers: Connect to multiple MCP servers simultaneously.
  • Add servers with unique names.
  • Manage (add/remove) servers via the UI.
  • Connect to all configured servers at the same time.

Available Tools

• View all available tools from the connected MCP server in the sidebar.
• Each tool displays:
  • Name and description.
  • Required and optional parameters and their types.
  • Parameter descriptions and constraints.
• Tools are automatically available to the LLM agent in the chat interface.
• Tool execution and its results are tracked in the chat history.

Future Improvement Plans

• STDIO MCP Server: Support connecting to the MCP server via standard input/output (STDIO) for more flexible server configuration.
• Individual Tool Testing: Implement the ability to test each tool individually in the UI to ensure proper functionality.
• ✅ Use Local LLMs: Support connecting to local LLMs (Llama, DeepSeek, Qwen...).
• Agent Memory: Introduce a memory function to allow the agent to retain context from interactions.
• RAG (File Upload): Enable Retrieval Augmented Generation (RAG) to allow users to upload files for the agent to enhance its responses.

Troubleshooting

• Connection Issues: Ensure that your MCP server is running and accessible.
• API Key Errors: Verify that you have entered the correct API keys for the LLM providers.
• Tool Errors: Check the server logs for detailed error information when using tools.

Important Note

This project is experimental and aims to explore and demonstrate how to use LangChain technology to interact with MCP servers. In practical applications, please adjust and optimize according to specific requirements.