Key Takeaways
- MetaGPT is free and open-source; the GitHub API reports
license.spdx_id: "MIT" - Canonical repository: github.com/FoundationAgents/MetaGPT, created June 30, 2023, written in Python
- 70,336 GitHub stars and 8,942 forks as of September 12, 2026, per the GitHub API
- The older
github.com/geekan/MetaGPTURL now redirects (HTTP 301) toFoundationAgents/MetaGPTβ both names refer to the same project - Installs with
pip install --upgrade metagpt, requiring Python 3.9 or later but below 3.12, plus Node.js and pnpm for full functionality - Core philosophy stated in the project's own README:
Code = SOP(Team)β role-based agents (product manager, architect, project manager, engineer) follow a defined Standard Operating Procedure - A separate Data Interpreter role (
metagpt.roles.di.data_interpreter) targets data-analysis and code-execution tasks rather than full application builds - Per the project's own docs/NEWS.md, the MetaGPT research paper was accepted for oral presentation (described there as "top 1.2%") at ICLR 2024, ranking #1 in the LLM-based Agent category β this is the project's own claim, not independently re-verified by PromptQuorum against ICLR's official program
π In One Sentence
MetaGPT is a free, open-source (MIT-licensed) Python multi-agent framework that simulates a software company, assigning role-based LLM agents to a Standard Operating Procedure that turns a one-line requirement into a PRD, system design, task breakdown, and code.
π¬ In Plain Terms
Instead of one AI model trying to plan and write an entire app in one prompt, MetaGPT splits the work between separate agents playing specific jobs β a product manager writes the requirements, an architect designs the system, an engineer writes the code β the same way a real software team would divide the work, just automated end to end.
πNote: This review is the single-subject companion to MetaGPT's entry in the Local LLM Software Directory β see that page for how MetaGPT compares at a glance to dozens of other local and hybrid AI tools. For a side-by-side look at MetaGPT alongside two unrelated apps, see the separate multi-agent frameworks comparison.
What Is MetaGPT?
MetaGPT is an open-source Python framework that simulates a software company made of role-based LLM agents. Its GitHub repository describes it with the tagline "The Multi-Agent Framework: First AI Software Company, Towards Natural Language Programming." Rather than one model generating an entire application from a single prompt, MetaGPT assigns a product manager, architect, project manager, and engineer role to separate agents, then routes work between them through a defined Standard Operating Procedure (SOP) β the project states its core philosophy as Code = SOP(Team).
- Core function: takes a one-line natural-language requirement and outputs artifacts including user stories, competitive analysis, requirements documents, data structures, APIs, and runnable code
- Role-based architecture: separate agents play product manager, architect, project manager, and engineer roles, each contributing the artifact that role would produce on a real team
- SOP-driven orchestration: agents hand off work through a fixed process rather than an unstructured, free-form conversation between agents
- Data Interpreter: a distinct role (
DataInterpreter) built for data-analysis and code-execution tasks β for example, running an analysis on a dataset and producing a plot β separate from the full-application software-company workflow - Canonical repository: github.com/FoundationAgents/MetaGPT, created June 30, 2023, written in Python, MIT-licensed per the GitHub API
What Can You Do With MetaGPT?
MetaGPT's feature set centers on turning a short requirement into a structured software-development pipeline, per the project's own README and documentation.
- One-line-to-repo generation β run
generate_repo("Create a 2048 game")(or the equivalent CLI command) and MetaGPT produces a project repository, walking the requirement through its role-based SOP - CLI usage β
metagpt "Create a 2048 game"creates a repository directly in a local./workspacedirectory, no Python script required - Library usage β
from metagpt.software_company import generate_repolets a Python application call MetaGPT programmatically and inspect the resultingProjectRepoobject - Data Interpreter β a dedicated role (
from metagpt.roles.di.data_interpreter import DataInterpreter) for data-analysis workflows, such as running an analysis on a dataset and producing a chart, rather than building a full application - Configurable model backend β a
~/.metagpt/config2.yamlfile (created viametagpt --init-config) sets the LLM provider, model name, base URL, and API key; the project's own example config lists OpenAI-compatible, Azure, and Ollama-style entries among supportedapi_typevalues - Example use cases documented by the project β data interpretation, multi-agent debate, an automated researcher role, and a receipt-processing assistant are named in the project's own tutorial index
How to Install and Get Started With MetaGPT
MetaGPT is a framework you install with pip, not a downloadable end-user app β there is no installer or app-store listing. The steps below follow the project's own README "Get Started" section; always check that page directly before installing, since exact commands can change between releases.
- 1Confirm your Python version is 3.9 or later but below 3.12 (
python --version) β the project documents this specific range as a requirement, and aconda create -n metagpt python=3.9 && conda activate metagptenvironment is one documented way to satisfy it. - 2Install the package:
pip install --upgrade metagpt(alternatives documented in the README include installing directly from GitHub withpip install --upgrade git+https://github.com/FoundationAgents/MetaGPT.git, or cloning the repository and runningpip install --upgrade -e .). - 3
- 4Initialize configuration: run
metagpt --init-configto create~/.metagpt/config2.yaml, then edit it to set your LLM provider'sapi_type,model,base_url, andapi_key. - 5Run it: from the command line,
metagpt "Create a 2048 game"generates a project in a local./workspacedirectory; from Python,from metagpt.software_company import generate_repoand callgenerate_repo("Create a 2048 game")to use it as a library instead.
Do I need an OpenAI account to run MetaGPT?
MetaGPT itself does not host or resell model access β you configure whichever LLM provider you want in ~/.metagpt/config2.yaml. The project's example config lists OpenAI-compatible, Azure, and Ollama-style api_type entries, so a locally hosted model reachable over an OpenAI-compatible API is a documented option, not just cloud providers.
Why does MetaGPT need Node.js and pnpm if it is a Python framework?
The project's own README states Node.js and pnpm must be installed "before actual use," in addition to the Python package β this review did not find a detailed public explanation of which specific MetaGPT feature depends on the Node.js toolchain, so treat it as a documented prerequisite rather than an optional step.
MetaGPT License and Pricing
The MetaGPT framework itself is free and open-source under the MIT License, confirmed via the GitHub API (license.spdx_id: "MIT"). There is no paid tier, seat license, or usage cap on the open-source framework; your only ongoing cost is usage of whichever LLM API provider you connect it to.
Component | Price | Notes |
|---|---|---|
| MetaGPT framework | Free (MIT) | Open source, self-hosted, no usage cap |
| Model API calls | Billed by provider | OpenAI, Azure, or a self-hosted/local model, depending on your config |
| MGX (MetaGPT X) | Separate commercial product | Hosted at mgx.dev, priced separately |
Verify current terms directly against the LICENSE file in the repository and, if considering MGX, against mgx.dev directly β this review covers the open-source framework, not MGX's commercial pricing or terms.
Who Should Use MetaGPT?
MetaGPT fits teams and individual developers who want a structured, role-based agent pipeline for generating an initial codebase from a short requirement, rather than a free-form multi-agent conversation framework.
MetaGPT vs. Alternatives
MetaGPT is one of several multi-agent frameworks in the agent-orchestration space, each built around a different coordination philosophy. Here is how its SOP-based, software-company-simulation approach compares to the closest alternatives β see the Local LLM Software Directory for the full catalog.
CrewAI
- Coordination Model:
- Role-based "crews" with defined tasks and a process (sequential or hierarchical)
- How It Differs From MetaGPT:
- Closest philosophical match β both use named roles and defined process, but CrewAI is a general-purpose crew abstraction for any task, not a fixed software-company SOP with PM/architect/engineer roles baked in.
AutoGen
- Coordination Model:
- Free-form, conversation-driven multi-agent chat with code-execution agents
- How It Differs From MetaGPT:
- AutoGen agents converge on a task through open-ended dialogue; MetaGPT instead routes agents through a fixed SOP with defined handoffs and output artifacts (PRD, design docs) at each stage.
LangGraph
- Coordination Model:
- Graph-based orchestration with explicit nodes, edges, branching, and state
- How It Differs From MetaGPT:
- LangGraph is a low-level orchestration engine you use to build any agent workflow, including a MetaGPT-like one; MetaGPT instead ships a pre-built software-company workflow rather than a graph-building toolkit.
AutoGPT
- Coordination Model:
- Single autonomous agent pursuing an open-ended goal with a task loop
- How It Differs From MetaGPT:
- AutoGPT centers on one agent looping toward a goal with minimal role structure; MetaGPT splits the work across multiple specialized roles following a fixed process instead of one agent iterating alone.
This is not an exhaustive list of multi-agent frameworks β see the Local LLM Software Directory for the full, regularly updated catalog, including MetaGPT's own directory entry.
Common Mistakes When Evaluating MetaGPT
Most confusion about MetaGPT comes from conflating it with the commercial MGX product, or treating self-reported project claims as independently verified facts.
Frequently Asked Questions
What is MetaGPT?
MetaGPT (github.com/FoundationAgents/MetaGPT) is a free, open-source Python multi-agent framework that simulates a software company, assigning product manager, architect, project manager, and engineer roles to LLM agents that turn a one-line requirement into a PRD, system design, and code.
Is MetaGPT free?
Yes. The framework is MIT-licensed and free, confirmed via the GitHub API license field. You separately pay usage costs to whichever LLM API provider you connect it to; MetaGPT itself charges nothing.
How do I install MetaGPT?
Run pip install --upgrade metagpt with Python 3.9 or later but below 3.12 installed. The project's README also documents installing directly from GitHub or from a local clone. Node.js and pnpm are additionally required "before actual use," per the README.
What is the Data Interpreter in MetaGPT?
The Data Interpreter is a dedicated role (metagpt.roles.di.data_interpreter.DataInterpreter) for data-analysis and code-execution tasks, such as running an analysis on a dataset and producing a plot β distinct from MetaGPT's full software-company, application-generation workflow.
Did MetaGPT really receive an ICLR 2024 oral presentation?
The project's own docs/NEWS.md states the MetaGPT paper was accepted for oral presentation ("top 1.2%") at ICLR 2024, ranking #1 in the LLM-based Agent category. This is the project's own claim; PromptQuorum reports it as self-reported rather than independently re-verified against ICLR's official published program.
What is MGX, and is it the same as MetaGPT?
MGX (MetaGPT X, mgx.dev) is a separate, commercial hosted product from the same project lineage, described by the project as launching in February 2025. It is not the open-source framework covered in this review, and its pricing and terms are set independently.
What license does MetaGPT use?
The MIT License, confirmed via the GitHub API (license.spdx_id: "MIT") β a permissive open-source license with no copyleft obligation.
How many GitHub stars does MetaGPT have?
MetaGPT's repository showed 70,336 stars and 8,942 forks as of September 12, 2026, verified directly via the GitHub API. Check the repository directly for a current count, since it changes daily.
Is geekan/MetaGPT the same repository as FoundationAgents/MetaGPT?
Yes. The older github.com/geekan/MetaGPT URL now returns an HTTP 301 redirect to github.com/FoundationAgents/MetaGPT, confirmed via the GitHub API β both names point to the same codebase and history.
How is MetaGPT different from CrewAI or AutoGen?
MetaGPT ships a fixed, software-company Standard Operating Procedure with named roles (product manager, architect, engineer) built in. CrewAI offers a more general-purpose role-based "crew" abstraction for any task, and AutoGen favors free-form, conversation-driven agent dialogue rather than a fixed SOP. See the comparison table above for details.