← Back to Discover

OpenManus

Python project for running a general-purpose AI agent from the terminal, with browser control through MCP, an MCP tool mode, an unstable multi-agent flow and an optional data analysis agent.

Autonomous agentsAgent frameworks
Visit website ↗

What it does

OpenManus is an open-source project from contributors who say they come from MetaGPT. Its README presents it as an agent anyone can run without an invite code, a reference to Manus, and describes it as a simple implementation whose prototype was built in a few hours. You install it with conda or uv, put your LLM settings (model, base URL and API key) in a config.toml file, and run python main.py to type an idea into the terminal. There is also an MCP tool version (run_mcp.py) and a multi-agent flow (run_flow.py) that the README calls unstable. Browser automation comes through a default Browser Use MCP server that attaches to a local Chrome or Chromium without needing an API key, with an option to use a Browser Use Cloud browser instead. An optional DataAnalysis agent can be switched on for data analysis and visualization tasks. The README also points to OpenManus-RL, a separate project on reinforcement-learning-based tuning of LLM agents.

Capabilities

  • Terminal-driven general agent started with python main.py
  • Your own LLM endpoint and key set in config.toml
  • Browser automation through a default Browser Use MCP server
  • MCP tool version and a multi-agent flow marked as unstable
  • Optional DataAnalysis agent for analysis and visualization

When to use it

  • You want an open-source general-purpose agent that you can run and modify yourself
  • You want to experiment with agents that drive a browser through MCP
  • You want to point an agent at the model provider of your choice

Alternatives to explore

  • Manus