Home / Tools / MetaGPT

MetaGPT

Multi-agent framework that writes software

Visit MetaGPT
MetaGPT preview
github.com Open Source

Upvotes

0

Pricing

Open Source

Category

Agents

In the database since

Oct 2026

Overview

About MetaGPT

MetaGPT is an open-source multi-agent framework that acts like a tiny software company. You give it one line of requirements, and agents playing product manager, architect, and engineer collaborate to produce user stories, APIs, documents, and code. The design follows careful standard procedures so agents stay organized. It installs with pip, supports many LLM providers, and is MIT-licensed with over 70,000 GitHub stars. The team also ships a separate commercial product called MGX. Its workflow research earned an oral spot at ICLR 2025.

Key Features

1. Role-based agent teams

Agents take software roles like product manager and engineer. Each role follows a standard procedure for its part of the job.

2. One-line requirement input

Describe the software you want in a single sentence. The team expands it into stories, designs, and working code.

3. Full deliverables

Output includes competitive analysis, data structures, APIs, documents, and code. You get the paperwork, not just a script.

4. Many LLM backends

Configure OpenAI, Azure, Ollama, Groq, and more through a YAML file. Swap models without changing your agents.

5. Simple install

Run pip install metagpt on Python 3.9 to 3.11. A Docker option exists for containerized setups.

6. Research pedigree

Its AFlow workflow research was an ICLR 2025 oral paper, ranking near the top of its category. The ideas are peer-reviewed.

Frequently Asked Questions (FAQ)

Where is the MetaGPT repo now?

The project moved from geekan/MetaGPT to FoundationAgents/MetaGPT on GitHub. Old links redirect to the new location.

What do I need to run it?

Python 3.9 through 3.11, plus Node.js and pnpm before real use. Run metagpt --init-config to set up your model keys.

What is MGX?

MGX, or MetaGPT X, is the team's separate commercial agent product. The framework itself stays MIT-licensed and free.

How good is the generated code?

It produces structured projects with docs, which beats raw snippets. Review everything, as with any AI-generated code.

Conclusion

MetaGPT turns the vague idea of agent teamwork into a concrete software process. The role-based design produces impressively complete project folders from tiny inputs. With 70,000 stars and active research behind it, the framework earns its reputation. Try it on a small tool first and read every file it writes. The new repo location is FoundationAgents.

Newsletter

Tools like this change fast

New Agents tools launch every week — the newsletter keeps you ahead of what's worth trying.

Keep exploring

Similar AI tools

Ranked by the job they do, their capabilities, and how closely they fit this workflow.