MetaGPT
MetaGPT is a Python multi-agent framework that models a software company: given a one-line requirement, a team of LLM agents acting as product manager, architect, project manager and engineer produces user stories, competitive analysis, requirements, data structures, API designs, documents and code. Its core idea is Code = SOP(Team), applying standard operating procedures to a team of agents.
It is best suited to researchers and developers exploring role-based multi-agent collaboration, and to anyone who wants to see a full idea-to-repository pipeline in action. Typical uses include generating a small project (for example `metagpt "Create a 2048 game"`), running the Data Interpreter for data analysis tasks, and studying the team's published research such as AFlow (an ICLR 2025 oral paper on automating agentic workflow generation). The team also runs a commercial product, MGX (MetaGPT X), built on these ideas.
Getting started: install with `pip install --upgrade metagpt` on Python 3.9 to 3.11, install Node.js and pnpm, then run `metagpt --init-config` and set your model provider and API key in `~/.metagpt/config2.yaml`. It works with OpenAI, Azure, Ollama, Groq and other providers.
Features
Why choose it
Trade-offs
Maintenance check
- Last commit
- 2026-01-21
- Commits, last 90 days
- 0
- Releases, last 90 days
- 0
- Latest release
- v0.8.2 (2025-03-09)
- Open issues
- 136
- Recent issues with at least one comment
- 13/17
Oldest week on the left; the latest four weeks are orange. Source: GitHub commit activity.
Status rule: archived on GitHub = Archived; last commit within 90 days = Active; 91 to 365 days = Slowing; older = Stale. All figures come from the GitHub API on the date shown.
Most discussed open issues
- #2110 Terminal.run_command() hangs indefinitely on Linux/WSL2 — end-of-command marker never yielded from splitlines/tuple-unpack5 comments · 2026-07-20
- #2156 Add Webz.io News Search as a search-engine provider1 comments · 2026-09-20
- #2153 Feature: verified knowledge transfer between role agents — memory bundles with proof and chain of custody1 comments · 2026-09-18
Issue titles are quoted from GitHub. Open the link to check whether a problem still applies to the version you use.
Who it fits
- You want to experiment with or demonstrate the 'software company SOP' multi-agent pattern (PM, architect, engineer generating specs and code).
- You are looking for the reference implementation of research papers from the team such as AFlow, SPO, and AOT.
- You want an actively updated general-purpose production agent framework; the repository's last push was in January 2026.
- You are on Python 3.12+ or do not want to install Node.js and pnpm alongside Python (README specifies Python 3.9–3.11).
How it differs
vs CrewAI★ 60.2kMetaGPT hard-codes a software-company SOP where agents produce PRDs, designs, and code; CrewAI is a general-purpose framework for custom role teams and event-driven workflows.
vs agentscope★ 34.4kMetaGPT focuses on SOP-driven software generation and research reproductions; AgentScope 2.0 is an actively maintained general toolkit for ReAct, multi-agent pipelines, and voice agents.
Compatibility
Quick start
Use cases
Alternatives
Featured in these lists
Comments
No comments yet. Be the first!