MetaGPT
StaleDescription
MetaGPT is a multi-agent framework that combines collaborative intelligence with software-engineering SOPs so agents cooperate like a real software company to ship code and docs.
Key Features
- Company-style multi-agent — built-in PM, architect, engineer and QA roles following real-world SOPs
- End-to-end output — a single requirement yields PRD, design, code and test cases
- Standardised pipeline — engineering flow from Message to Plan to Code
- Customisable roles — define new roles, tools and workflows via simple config
- Multi-language — generates Python, Java, JavaScript and other language code
- Open ecosystem — widely adopted in research and industry, with Data Interpreter and MGX extensions
Use Cases
💡 Turning a one-line product spec into a multi-agent MVP with code
💡 Standing up multi-agent software-engineering pipelines inside an organisation
💡 Encoding engineering experience as standardised Agent SOPs
💡 Combining with LangChain / LlamaIndex to build research-style multi-agent apps
💡 Using as a research framework for studying LLM multi-agent collaboration
Categories
Quick Start
# Install
pip install metagpt
# Configure an LLM (OpenAI / DeepSeek / any OpenAI-compatible endpoint)
export OPENAI_API_KEY=sk-xxx
# Kick off the multi-agent team
python -m metagpt.software_company 'Build a snake game'
# The generated project appears in the workspace/ directory