Multiagents Thinking

This project is an MCP server based on a multi - agent system. It analyzes and solves complex tasks by generating multiple agents with professional perspectives in parallel, supports multiple problem - solving strategies, and integrates the outputs of each agent to generate a final solution.

Artificial intelligence chatbots Education and learning tools #Multi - agent #Parallel computing #Task decomposition #MCP protocol .Python

rating : 2 points

downloads : 15

update time : 2025-04-29

What is the Multi - Agent Thinking MCP Server?

This is an intelligent task processing system that can automatically decompose complex problems into multiple subtasks and assign AI agents with different professional expertise to process them in parallel, and finally integrate the optimal solution. The system is particularly suitable for decision - making scenarios that require multi - angle analysis.

How to use this service?

Multi - agent collaboration can be initiated through simple natural language instructions. It supports direct calls from development tools such as Cursor IDE and can also be configured for use with AI assistants such as Claude Desktop.

Applicable scenarios

Complex decision - making scenarios that require multi - dimensional thinking, such as product design evaluation, business strategy analysis, technical solution comparison, and academic research.

Main functions

Multi - agent collaborationAutomatically generate 3 - 5 AI agents with different professional perspectives to process tasks in parallel

Diverse solution strategiesSupports 6 problem - solving modes such as perspective sharing, task decomposition, and stage collaboration

Intelligent result integrationAutomatically synthesize the analysis reports of each agent to generate a structured final solution

Advantages and limitations

Advantages

Provide a more comprehensive analysis perspective than a single AI

Reduce the processing time of complex tasks by 40% - 60%

Increase the diversity of solutions by more than 3 times

Limitations

Require OpenAI and HuggingFace API keys

Complex tasks may incur high API call costs

Currently mainly support English processing

How to use

Environment preparation

Install Python 3.11+ and the uv package manager, and obtain the API key

Configure environment variables

Create a .env file in the project root directory and add the API key

Start the server

Run the main program to start the MCP service

Usage examples

Education platform designAnalyze the design scheme of an online learning platform from three dimensions: pedagogy, technical implementation, and user experience

Market strategy evaluationAnalyze the marketing, legal, and logistics challenges of a new product entering the European market in parallel

Frequently Asked Questions

How much API call cost is required?

Does it support Chinese tasks?

How to monitor the working status of agents?

Related resources

Project GitHub repository

Source code and latest updates

smolagents documentation

Infrastructure technical documentation

FastMCP tutorial

MCP server configuration guide

🚀 Multi-Agent Thinking MCP Server

This project aims to achieve more in - depth and multi - perspective analysis and generate problem - solving strategies by generating multiple expert agents to share the processing of complex tasks. Multiple internally generated agents (referred to as expert agents) will be assigned and execute tasks according to their respective areas of expertise.

✨ Features

Multi - Agent Thinking

Automatically generate various different perspectives or methods from a single task description.
Run expert agents dedicated to each method in parallel.
Support multiple problem - solving strategies, such as perspective - allocation, task - allocation, phase - allocation, parallel - solution, branch - exploration, and decomposition - reconstruction types.
Integrate reports from each expert agent to generate a final conclusion or solution.

MCP Server

Implement the MCP server functionality using fastmcp.
The multiagent.multiagent_thinking tool can be invoked through MCP clients such as Cursor IDE.

📦 Installation

Requirements

Python 3.11 or higher
uv package manager
Necessary API keys:
- OpenAI API key (OPENAI_API_KEY)
- HuggingFace token (HF_TOKEN)

Steps

Clone the repository:

git clone https://github.com/Hajime-Y/multiagents-thinking.git
cd multiagents-thinking

Create a virtual environment and install dependencies:

uv venv
source .venv/bin/activate # Linux/macOS
# .venv\Scripts\activate # Windows
uv sync

Environment Variables

Create a .env file in the project root directory and set the following environment variables:

OPENAI_API_KEY=your_openai_api_key
HF_TOKEN=your_huggingface_token

💻 Usage Examples

Basic Usage

Start the MCP server:

uv run python -m src.multiagents

Invoke the multiagent_thinking tool in a client (e.g., Cursor IDE).

For the configuration of Claude Desktop, refer to the instructions in the documentation to ensure correct setting of commands and parameter paths.

📚 Documentation

Main Components

src/multiagents.py: The entry point of the MCP server, which defines the multiagent_thinking tool.
src/create_agent.py: Responsible for generating prompts for the mother agent and expert agents and the logic for agent creation (includes the run_specialist_agents tool).