# MetaGPT: features, protocols, quickstart

## TL;DR

MetaGPT is a Python multi-agent framework, now under the FoundationAgents GitHub organization, that models a software company: product manager, architect, project manager and engineer roles follow an SOP to turn a one-line requirement into documents and code. Activity has slowed, with the last push in January 2026. It suits researchers studying role-based agent teams.

## Key facts

| Field | Value |
| --- | --- |
| Type | Framework |
| Languages / SDKs | Python |
| License | MIT |
| Pricing model | Open source, free |
| Orchestration pattern | Crew / roles |
| GitHub stars | 70,709 (as of 2026-10-01) |
| GitHub forks | 8,988 |
| Last push | 2026-01-21 |
| Latest release | v0.8.1 |
| Repository | [FoundationAgents/MetaGPT](https://github.com/FoundationAgents/MetaGPT) |
| Documentation | [docs.deepwisdom.ai](https://docs.deepwisdom.ai/main/en/) |
| Last verified | 2026-09-30 |

## Key features

- Built-in software-company roles (ProductManager, Architect, ProjectManager, Engineer) hired into a Team and run for a set number of rounds. ([source](https://docs.deepwisdom.ai/main/en/guide/get_started/quickstart.html))
- Custom agents are Roles composed of Actions, with an observe-think-act loop. ([source](https://docs.deepwisdom.ai/main/en/guide/tutorials/agent_101.html))
- Roles subscribe to upstream message types with _watch and publish results to a shared environment. ([source](https://docs.deepwisdom.ai/main/en/guide/tutorials/multi_agent_101.html))
- Any role can be switched to a human participant with is_human=True, without changing the SOP. ([source](https://docs.deepwisdom.ai/main/en/guide/tutorials/human_engagement.html))
- Data Interpreter role that plans and writes code for data-analysis tasks. ([source](https://docs.deepwisdom.ai/main/en/guide/use_cases/agent/interpreter/intro.html))
- CLI options for incremental work on an existing repo (--inc) and recovering a run from serialized storage (--recover-path). ([source](https://docs.deepwisdom.ai/main/en/guide/get_started/quickstart.html))
- LLM provider set in ~/.metagpt/config2.yaml, with api_type values such as openai, azure, ollama and groq. ([source](https://docs.deepwisdom.ai/main/en/guide/get_started/configuration/llm_api_configuration.html))

## Architecture and orchestration pattern

Pattern: Crew / roles.

An agent is a `Role` that owns a list of `Action`s and loops through observe, think and act. Roles do not call each other directly: each one declares which message types it reacts to with `_watch`, and publishes its output as a `Message` into a shared environment that other roles observe.

A `Team` wraps that environment. You `hire` roles, set a dollar budget with `invest`, submit the requirement with `run_project`, and `run(n_round=...)` rounds. The bundled software-company SOP passes work from product manager to architect to project manager to engineer, with optional code review and QA.

Each role's memory is the list of messages it has observed (`rc.memory`), read back with `get_memories(k)`. Generated artifacts are written to a project workspace, and the CLI can resume a project from serialized storage or work incrementally on an existing repository.

### Human in the loop

Setting `is_human=True` on a role (for example a reviewer) hands that role to a person: when its turn comes, the run pauses for terminal input and the reply is passed to the other agents as that role's message. The docs state two limits: input is terminal-only, and the human must follow the same content format the downstream logic expects from an LLM. Custom `_act` code must call actions registered via `set_actions` for this to work. No approval gate for individual tool calls is documented.

## Protocols

| Protocol | Support | Evidence | Note |
| --- | --- | --- | --- |
| MCP | Unknown (checked 2026-09-30) | — | Searched README, docs.deepwisdom.ai guide pages, GitHub code search and issues for mcp / model context protocol; only third-party proposals in issues. |
| A2A | Unknown (checked 2026-09-30) | — | Searched README, docs pages, GitHub code search and issues for a2a / agent2agent; nothing official found. |
| AG-UI | Unknown (checked 2026-09-30) | — | Searched README, docs, GitHub code search for ag-ui / agui, and the AG-UI README integration list; MetaGPT is not listed. |

## Best for

- Generating a first-draft repository with PRD, design documents and code from a one-line requirement. ([shortlist](https://multiagentguide.top/best/coding-agents.md))
- Studying SOP-driven, role-based collaboration between LLM agents.
- Research-report agents built on the documented Researcher use case. ([shortlist](https://multiagentguide.top/best/research-agents.md))
- Mixing human and LLM participants in the same role pipeline.

## Not for

- Projects that must run on Python 3.12 or newer.
- Teams that need frequent releases or documented MCP / A2A support.
- Human approval flows that need a web UI rather than terminal input.

## Quickstart

```sh
pip install --upgrade metagpt
```

Install not yet verified by this site.

```python
import asyncio
from metagpt.roles import Architect, Engineer, ProductManager, ProjectManager
from metagpt.team import Team

async def main(idea: str):
    company = Team()
    company.hire([ProductManager(), Architect(), ProjectManager(), Engineer()])
    company.invest(investment=1.0)   # budget in US dollars for model calls
    company.run_project(idea=idea)
    await company.run(n_round=4)

# Needs ~/.metagpt/config2.yaml (create it with: metagpt --init-config)
asyncio.run(main("A command-line tool that converts CSV files to Markdown tables"))
```

### Common pitfalls

- The README requires Python 3.9 or later but below 3.12.
- Run `metagpt --init-config` and put the model key in `~/.metagpt/config2.yaml` before the first run.
- The README asks for Node.js and pnpm before actual use; diagram output needs a Mermaid engine such as `@mermaid-js/mermaid-cli`.
- The docs estimate about $0.2 of GPT-4 fees for an analysis-and-design example and about $2.0 for a full project (project's own estimate).
- Besides the PyPI package, the README lists installing from the Git repository (`pip install --upgrade git+https://github.com/geekan/MetaGPT.git`) or from an editable clone (`pip install --upgrade -e .`), and the installation docs describe a Docker image (`metagpt/metagpt:latest`).
- A closed issue (#1910) reports `pip install metagpt==0.8.2` failing dependency resolution through a newer `agentops`; the reporter's workaround was pinning `agentops==0.3.17`.
- GitHub marks v0.8.1 as the latest release although v0.8.2 was published in March 2025.

Install check by this site (2026-09-30, `uv venv --python 3.11 && uv pip install --upgrade metagpt`, macOS arm64) failed: dependency resolution fails. `metagpt` 0.8.2 (latest on PyPI, requires Python >=3.9,<3.12) pins `lancedb==0.4.0`, and no 0.4.x release of `lancedb` is available on pypi.org as of 2026-09-30, so the resolver cannot select 0.8.2; uv then backtracked to `metagpt` 0.1, whose `pandas` 1.4.1 source build failed. Tested with Python 3.11.

Official quickstart: https://docs.deepwisdom.ai/main/en/guide/get_started/installation.html

## Pros

- A full software-company SOP ships in the box, producing user stories, requirements, API design and code from one prompt. ([source](https://docs.deepwisdom.ai/main/en/guide/get_started/introduction.html))
- A human can take over any role with a single flag while the rest of the pipeline stays unchanged. ([source](https://docs.deepwisdom.ai/main/en/guide/tutorials/human_engagement.html))
- Explicit spending control: the team is given a dollar budget with invest before it runs. ([source](https://docs.deepwisdom.ai/main/en/guide/tutorials/multi_agent_101.html))
- Supports many model backends through config2.yaml, including local models via Ollama. ([source](https://docs.deepwisdom.ai/main/en/guide/get_started/configuration/llm_api_configuration.html))
- Incremental mode and serialized recovery let a generated project be extended or resumed. ([source](https://docs.deepwisdom.ai/main/en/guide/get_started/quickstart.html))

## Cons

- Python support stops below 3.12, per the README installation note. ([source](https://github.com/FoundationAgents/MetaGPT))
- Development has slowed: the last tagged release is v0.8.2 (March 2025) and the last push to main was in January 2026. ([source](https://github.com/FoundationAgents/MetaGPT/commits/main))
- Human participation is terminal-only and requires the person to match the format expected from an LLM, as the docs note. ([source](https://docs.deepwisdom.ai/main/en/guide/tutorials/human_engagement.html))
- Installing the latest PyPI release has dependency-resolution problems: a closed issue reports pip failing on metagpt==0.8.2 until agentops is pinned to 0.3.17. ([source](https://github.com/FoundationAgents/MetaGPT/issues/1910))
- The README's most recent news items (2025) cover the team's hosted MGX product and papers rather than framework releases. ([source](https://github.com/FoundationAgents/MetaGPT))

## Alternatives

- [ChatDev](https://multiagentguide.top/tools/chatdev.md) ([MetaGPT vs ChatDev](https://multiagentguide.top/compare/metagpt-vs-chatdev.md))
- [CrewAI](https://multiagentguide.top/tools/crewai.md)
- [CAMEL](https://multiagentguide.top/tools/camel.md)
- [OpenManus](https://multiagentguide.top/tools/openmanus.md)
- [AutoGen](https://multiagentguide.top/tools/autogen.md)

## FAQ

### Is MetaGPT still maintained?

The repository is not archived, but activity is low: the last push was on 2026-01-21 and the newest tagged release is v0.8.2 from March 2025.

### Does MetaGPT support MCP?

No official MCP, A2A or AG-UI support was found in the README, docs or code; all three are recorded as unknown.

### Is MetaGPT free?

Yes. The framework is MIT-licensed with no paid tier of its own; you pay your model provider. The same team sells MGX (mgx.dev now redirects to Atoms at atoms.dev), which its site presents as an AI website and app builder with free and paid plans rather than as a hosted MetaGPT service.

### What Python version does MetaGPT need?

Python 3.9 or later but below 3.12, according to the README.

### How do MetaGPT agents coordinate?

Each role watches for specific upstream message types and publishes its output to a shared environment, following an SOP; a Team sets the budget and number of rounds, and a human can replace any role via is_human=True.

## Sources

- [FoundationAgents/MetaGPT repository (README)](https://github.com/FoundationAgents/MetaGPT)
- [Commit history (main)](https://github.com/FoundationAgents/MetaGPT/commits/main)
- [Release v0.8.2](https://github.com/FoundationAgents/MetaGPT/releases/tag/v0.8.2)
- [MetaGPT documentation](https://docs.deepwisdom.ai/main/en/)
- [Introduction](https://docs.deepwisdom.ai/main/en/guide/get_started/introduction.html)
- [Installation](https://docs.deepwisdom.ai/main/en/guide/get_started/installation.html)
- [Quickstart](https://docs.deepwisdom.ai/main/en/guide/get_started/quickstart.html)
- [LLM API configuration](https://docs.deepwisdom.ai/main/en/guide/get_started/configuration/llm_api_configuration.html)
- [Agent 101](https://docs.deepwisdom.ai/main/en/guide/tutorials/agent_101.html)
- [MultiAgent 101](https://docs.deepwisdom.ai/main/en/guide/tutorials/multi_agent_101.html)
- [Human engagement](https://docs.deepwisdom.ai/main/en/guide/tutorials/human_engagement.html)
- [Use memories](https://docs.deepwisdom.ai/main/en/guide/tutorials/use_memories.html)
- [Concepts](https://docs.deepwisdom.ai/main/en/guide/tutorials/concepts.html)
- [Data Interpreter use case](https://docs.deepwisdom.ai/main/en/guide/use_cases/agent/interpreter/intro.html)
- [Researcher use case](https://docs.deepwisdom.ai/main/en/guide/use_cases/agent/researcher.html)
- [Atoms (MGX) pricing](https://atoms.dev/pricing)
- [Issue #1910: pip install metagpt==0.8.2 resolution failure (agentops pin)](https://github.com/FoundationAgents/MetaGPT/issues/1910)

## Unknown fields

protocols.mcp, protocols.a2a and protocols.agui are unknown: the README, docs.deepwisdom.ai guide pages, GitHub code search and issue search found no official support (only third-party integration proposals), and the AG-UI integration list does not include MetaGPT. The GitHub snapshot's latest_release is v0.8.1 because GitHub marks it as latest; v0.8.2 exists.

Corrections or removal requests: support@multiagentguide.top

---

Data as of 2026-10-01. Not affiliated with listed projects. HTML version: https://multiagentguide.top/tools/metagpt