MCP LLM client

Python client, based on "mcp[cli]", for connecting to MCP servers through multiple protocols, specifically designed to work with integrated language models.

Table of Contents

• Overview
• Installation
• Implemented Models
• Implemented Transfer Protocols
• System Requirements
• Usage Example
• Version History
• Project Status
• License

Overview

This package provides a Python interface to connect to MCP servers in an easy, intuitive, and configurable way. It offers a modular architecture that allows for easy extension of new transfer protocols and language models. Currently includes support for HTTPStream and GPT-4 mini, with expansion capability for more options in the future.

Installation

To install the MCP client, you can use pip:

pip install mcp-llm-client

Implemented Models

The client currently supports the following language models:

🚨 CRITICAL CONFIGURATION NOTE Currently, this project only work with gpt4o-mini llm model.

Implemented Transfer Protocols

Protocols for communication with MCP servers:

🚨 CRITICAL CONFIGURATION NOTE Currently, this project only work with HTTPStream protocol.

Future Development Planning

Pending Language Models

• Integration of additional language models
• Implementation of dynamic model selection system
• Optimization of model loading and management

Pending Protocols

• Complete implementation of SSE for better real-time event handling
• Development of stdio interface for local environments
• Performance optimization across all protocols

System Requirements

Environmental Configuration

• .env file: The .env file contains the authentication credentials necessary for integration with external services. This file must be created in the project root directory with the following format:

  # .env
  # OpenAI Authentication
  OPENAI_API_KEY=<YOUR OPENAI-API-KEY>
  

• config.json file: The config.json file defines the configuration of available MCP servers. It must be created in the project root directory with the following structure:

  {
      "app_name": "mcp-llm-client",
      "mcp_servers": {
          "example_public_server": {
              "transport": "httpstream",
              "httpstream-url": "http://127.0.0.1:8000/public-example-server/mcp",
              "name": "example-public-server",
              "description": "Example public server."
          },
          "example_private_mcp": {
              "transport": "httpstream",
              "httpstream-url": "http://127.0.0.1:8000/private-example-server/mcp",
              "name": "example-private-server",
              "description": "Example private server with oauth required.",
              "auth": {
                  "required": true,
                  "post_body": {
                      "username": "user",
                      "password": "password"
                  }
              }
          },
          "github": {
              "transport": "httpstream",
              "httpstream-url": "https://api.githubcopilot.com/mcp",
              "name": "github",
              "description": "This server specializes in github operations.",
              "auth": {
                  "required": false,
                  "post_body": null
              },
              "headers": {
                  "Authorization": "Bearer {access_token}"
              }
          }
      }
  }
  

  If you need an MCP server to test the code, you can use simple-mcp-server.

Dependencies

• Python = ">=3.11"
• openai = "^1.68.2"
• mcp[cli]
• mcp-oauth

Usage Example

#example1.py
from mcpclient import open_local_chat
open_local_chat()

#example2.py
from mcpclient import Chat
chat: Chat = Chat()
while True:
    query = input("> ")
    if query == "":
        break
    for step in chat(query):
        print(f"<< {step.json}")

Alternatively, you may test this service using the following template available on GitHub:

# clone repo
git clone https://github.com/rb58853/template_mcp_llm_client.git

# change to project dir
cd template_mcp_llm_client

# install dependencies
pip install -r requirements.txt

# open in vscode
code .

Version History

Last Version Features

• 💬 Fully functional chat by passing a query; see Chat.
• ⚙️ Integration with Tools, Resources, and Prompts from MCP servers, achieving a well-integrated client workflow with each of these services.
• 🔐 Simple authentication system using mcp-oauth and this environmental configuration. Also integrate headers authorization.
• 👾 OpenAI GPT as an integrated LLM using the model "gpt4o-mini".
• 📡 Support for the httpstream transport protocol.
• 💻 Easy console usage via open_local_chat(); see example1 for the use case.

See more in changelog

Project Status

⚠️ Important Notice: This project is currently in active development phase. As a result, errors or unexpected behaviors may occur during usage

License

MIT License. See license.