Skip to content
← Tools / Skills
ToolDeepWisdom (FoundationAgents)/10 trustVerified

MetaGPT

Multi-agent framework that runs a virtual software company (PM → engineer).

Visit official site →

About

MetaGPT is the multi-agent framework famous for simulating a whole software company: give it a one-line requirement and role-based agents (product manager, architect, project manager, engineer) collaborate through structured SOPs to produce user stories, APIs, and code. 70.8k stars, MIT, with ICLR-published research (AFlow oral, top 1.8%). Maintenance note: the repo was last pushed January 2026 (9 months) with 0 releases in 12 months — the team is now focused on the commercial MGX product (mgx.dev) — so treat the OSS framework as maintenance-mode and verify activity before building on it. Now under the FoundationAgents org (moved from geekan; old URLs redirect).

What it can do

  • **Role-based agents**: PM, architect, engineer, QA with SOP-driven collaboration
  • **One-line to repo**: requirements → stories → APIs → code
  • **Data Interpreter**: code-writing agent for data tasks
  • **Multi-agent patterns**: debate, researcher, and custom roles
  • **Research-backed**: ICLR 2024 paper + follow-up agent research

Setup

```bash

pip install metagpt # Python 3.9–3.11 required

metagpt --init-config # then set your LLM API key in ~/.metagpt/config2.yaml

```

```bash

metagpt "Create a 2048 game" # generates a repo in ./workspace

```

Also needs node + pnpm for full use. Docs: docs.deepwisdom.ai.

Prerequisites

Python 3.9–3.11 (NOT 3.12+); node + pnpm. An LLM API key (OpenAI/Azure/Ollama/Groq).

Use cases

multi-agent systemscode generationresearchSOPs

Works with

customGeneric API

Trust score — 8/10

/10 trustVerified

Scored 8/10 (verified tier, metrics-v1). Evidence: 70,821 GitHub stars (log-scaled vs catalog peers); last push 264d ago; 0 releases/12mo; 141 open issues. Manual review: famous software-company-of-agents (70.8k stars, ICLR-published); BUT pushed Jan 2026 (9mo), 0 releases/12mo — team focused on MGX product; fiddly setup (py<3.12, node/pnpm). License: MIT.

★ 70,821 GitHub stars

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
9.0/10
Maintenance · recency, releases, issue load
4.8/10
Openness · license + docs
10.0/10
Manual review · sanity read + red-flag check
7.0/10
  • · 70,821 GitHub stars (log-scaled vs catalog peers)
  • · last push 264d ago; 0 releases/12mo; 141 open issues
  • · manual review: famous software-company-of-agents (70.8k stars, ICLR-published); BUT pushed Jan 2026 (9mo), 0 releases/12mo — team focused on MGX product; fiddly setup (py<3.12, node/pnpm)

Sources

Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.