Aworld

v1.0.0Coding Agentsstable

Search, understand, reproduce, and improve an idea with ease

agent-frameworkagent-learningagent-runtimebrowsecompenvironment
Share:
1,198
Stars
0
Downloads
0
Weekly
0/5

What is Aworld?

Aworld is a Model Context Protocol (MCP) server that allows AI assistants like Claude, Cursor, and VS Code to search, understand, reproduce, and improve an idea with ease

Search, understand, reproduce, and improve an idea with ease

This server falls under the Coding Agents category on MCPgee, the world's largest MCP server directory with 33,000+ servers.

Features

  • Search, understand, reproduce, and improve an idea with ease

Use Cases

Search, understand, and improve ideas with ease.
Use agent learning and RL training for optimization.
inclusionAI

Maintainer

LicenseMIT
Languagepython
Versionv1.0.0
UpdatedMay 21, 2026
Statushealthy
Maintenanceactive

Works with

ClaudeOpenAIwindowsmacoslinux

Installation

Manual Installation

npx aworld

Configuration

Configuration Details

Config File

claude_desktop_config.json

Performance

Response Metrics

Response Time< 200ms
ThroughputMedium

Resource Usage

Memory UsageLow
CPU UsageLow

How to Set Up and Use Aworld

AWorld is an open-source multi-agent framework and MCP server that provides an environment for AI agents to search, understand, reproduce, and iteratively improve ideas and software projects. It orchestrates specialized agents — including a Developer Agent for coding, an Evaluator Agent for quality assessment, and media generation agents for video, audio, and images — within an automated evolution loop that builds, evaluates, and refines outputs until quality targets are met. Researchers and developers working on automated software generation, RL-based agent training, and benchmark evaluation (GAIA, BrowseComp, xBench) use AWorld as their agent runtime.

Prerequisites

  • Python 3.11+ installed (the framework requires 3.11 specifically)
  • Conda package manager for environment isolation
  • An MCP-compatible AI client such as Claude Desktop or Claude Code
  • API keys for the LLM providers you want to use (configured via .env file)
  • Git, to clone the AWorld repository
1

Clone the repository and create a conda environment

Clone AWorld from GitHub and create an isolated Python 3.11 conda environment. The framework requires Python 3.11 specifically.

git clone https://github.com/inclusionAI/AWorld && cd AWorld
conda create -n aworld_env python=3.11 -y
conda activate aworld_env
2

Install AWorld and the CLI

Install the main AWorld package and the aworld-cli tool in development mode so you have access to all commands.

pip install -e .
cd aworld-cli && pip install -e .
3

Configure your LLM provider and API keys

Run the interactive CLI configuration to set up your model provider and API credentials. Alternatively, create a .env file in your working directory with the required environment variables.

aworld-cli --config
4

Start the AWorld MCP server

Launch AWorld as an MCP server so your AI client can delegate tasks to the multi-agent system. The server exposes the framework's agent capabilities as MCP tools.

aworld-cli --mcp
5

Configure your AI client

Add the AWorld MCP server to your claude_desktop_config.json so Claude or another MCP client can invoke AWorld's multi-agent capabilities.

{
  "mcpServers": {
    "aworld": {
      "command": "aworld-cli",
      "args": ["--mcp"]
    }
  }
}
6

Run an evolution loop task

Ask the AI assistant to use AWorld to build and iteratively improve a project. The evolution loop will automatically delegate to the Developer Agent, score the output with the Evaluator Agent, and refine until quality targets are met.

Aworld Examples

Client configuration

Claude Desktop configuration for the AWorld MCP server using the aworld-cli command.

{
  "mcpServers": {
    "aworld": {
      "command": "aworld-cli",
      "args": ["--mcp"],
      "env": {
        "OPENAI_API_KEY": "your-openai-api-key"
      }
    }
  }
}

Prompts to try

Use these prompts in your AI client once the AWorld MCP server is running and configured.

- "Use AWorld to create a Python web scraper that extracts product prices from an e-commerce site"
- "Start an evolution loop to build a data visualization dashboard with a UI quality score above 0.85"
- "Search for and reproduce the algorithm described in this research paper abstract"
- "Use the Developer Agent to refactor my codebase and then have the Evaluator Agent score the improvements"

Troubleshooting Aworld

aworld-cli command not found after installation

Ensure you have activated the conda environment ('conda activate aworld_env') before running aworld-cli. The CLI is installed in the environment's bin directory and will not be available in other environments or system Python.

Evolution loop fails without completing the task

Check your .env file to ensure LLM API keys are set correctly and the configured model is accessible. The evolution loop requires multiple sequential LLM calls; if a call fails mid-loop, the loop will abort. Increase API rate limits or switch to a model with higher throughput.

Agent generates code but evaluation scores remain low

The Evaluator Agent compares output against the stated quality criteria. Be more specific in your prompt about what 'quality' means (e.g., 'the UI must load in under 2 seconds and pass WCAG AA accessibility checks'). More concrete criteria lead to more targeted improvement cycles.

Frequently Asked Questions about Aworld

What is Aworld?

Aworld is a Model Context Protocol (MCP) server that search, understand, reproduce, and improve an idea with ease It connects AI assistants to external tools and data sources through a standardized interface.

How do I install Aworld?

Follow the installation instructions on the Aworld GitHub repository. Clone the repo, install dependencies, and add the server config to your AI client.

Which AI clients work with Aworld?

Aworld works with all major MCP-compatible AI clients including Claude Desktop, Claude Code, Cursor, VS Code (GitHub Copilot), Windsurf, and Cline.

Is Aworld free to use?

Yes, Aworld is open source and available under the MIT license. You can use it freely in both personal and commercial projects.

Browse More Coding Agents MCP Servers

Explore all coding agents servers available in the MCPgee directory. Each server includes setup guides for Claude, Cursor, and VS Code.

Quick Config Preview

{ "mcpServers": { "aworld": { "command": "npx", "args": ["-y", "aworld"] } } }

Add this to your claude_desktop_config.json or .cursor/mcp.json

Read the full setup guide →

Ready to use Aworld?

Browse our complete directory of 33,000+ MCP servers, read setup guides for your editor, and start building with the Model Context Protocol.

33,000+ ServersFree & Open SourceStep-by-Step Guides