Multi-Agent Thinking MCP Server

@Hajime-Y

Basic information

About Multi-Agent Thinking MCP Server

MCP server for multi-agent thinking using smolagents

Multi-Agent Thinking MCP Server config

Add this server to your MCP-compatible client using the configuration below.

{
  "mcpServers": {
    "multiagents-thinking": {
      "command": "uv",
      "args": [
        "venv"
      ]
    }
  }
}

Multi-Agent Thinking MCP Server tools

Tool list not available yet

This listing does not currently have a tool list to display. This does not mean the server has no tools; check its documentation for supported capabilities.

We auto-extract tools from the README. The maintainer can list them under a ## Tools heading to populate this section.

Overview

What is Multi-Agent Thinking MCP Server?

Multi-Agent Thinking MCP Server is a server that generates multiple specialist agents to tackle complex tasks from different perspectives and approaches in parallel. It is built on HuggingFace’s smolagents library and FastMCP, and is designed for users who need deep, multi‑angle analysis or solution generation.

How to use Multi-Agent Thinking MCP Server?

Clone the repository, create a virtual environment with uv, install dependencies with uv sync, then set OPENAI_API_KEY and HF_TOKEN in a .env file. Start the server with uv run python -m src.multiagents. MCP clients such as Cursor IDE can invoke the tool via multiagent.multiagent_thinking. For Claude Desktop, add a configuration entry in the MCP servers settings.

Key features of Multi-Agent Thinking MCP Server

Automatically generates multiple perspectives from a single task description.
Runs specialist agents in parallel, each focusing on a distinct approach.
Supports diverse problem‑solving strategies: perspective sharing, task sharing, stage sharing, parallel solving, branch search, and decomposition‑reconstruction.
Integrates reports from all specialists into a final conclusion or solution.

Use cases of Multi-Agent Thinking MCP Server

Analyze a design problem from several disciplinary viewpoints simultaneously.
Generate competing solution approaches for a complex task.
Decompose a large problem into sub‑tasks handled by separate specialist agents.

Frequently asked questions

What are the runtime requirements?

Python 3.11 or later, the `uv` package manager, and API keys for OpenAI and HuggingFace.

What API keys are needed?

You must provide `OPENAI_API_KEY` and `HF_TOKEN` in a `.env` file at the project root.

How do I configure Multi-Agent Thinking MCP Server for Claude Desktop?

Add a JSON entry to your Claude Desktop configuration (e.g., `~/.config/Claude/claude_desktop_config.json`) with the absolute path to `uv` and to the project directory, using `uv run python -m src.multiagents` as the command.

How do I start the server?

Run `uv run python -m src.multiagents` from the project root after installing dependencies and setting environment variables.

What tool does the server expose?

It exposes the `multiagent.multiagent_thinking` tool for MCP clients to call.

Comments

More AI & Agents MCP servers

XDC AI

XDC AI is a remote MCP server available at https://xdcai.tech.

InsumerAPI

Wallet auth for AI agents: send a blockchain wallet and conditions, get a signed yes or no across 37 chains, never the balance. Hosted at https://api.insumermodel.com/mcp with no key (10 tools), or run locally with npx -

scvd.store

seancrecord

scvd.store is an evidence observatory for agentic commerce. Independent signed observation of what other people's endpoints, artifacts and payments actually did among other things. USDC on Base, Arbitrum One, World Chain

Gamedev Tavern

Gamedev Tavern is MCP for the game industry: studios, people, games and open quests. Have a goal — your AI agent writes a quest in your words and Tavern agents look for people and studios that fit. No goal — your agent i

Phi Longevity PRISM

Phi Longevity

Guideline-cited lab-results analysis for people managing type 2 diabetes, kidney disease, lupus and cancer survivorship. Flags tests that are missing or overdue (ADA, KDIGO, NCCN — each cited) and returns questions to br