modelscope-agent
MS-Agent is a lightweight framework for building agents with autonomous exploration, supporting MCP-based tool calling, deep research, code generation, and video generation. It solves the problem of creating complex, multi-capability agents for general-purpose tasks. It is for developers who want to build and customize AI agents.
✨ Key features
- Multi-agent chat with MCP tool calling
- Deep research for autonomous exploration
- Code generation with artifacts
- Short video generation (~5 minutes)
- Agent skills with progressive disclosure
- WebUI with SSE-streamed interaction
🎯 Use cases
- Build a chatbot that can call external tools via MCP
- Automate deep research and report generation
- Generate code for complex programming tasks
- Create short videos from prompts
- Develop financial research workflows with multi-agent
📦 Installation
🧰 Requirements: Python environment; API keys for LLMs may be required depending on the model used.
pip install ms-agentpip install 'ms-agent[research]'Or install from source:
git clone https://github.com/modelscope/ms-agent.git
cd ms-agent
pip install -e .
🚀 Usage
# Example from README (truncated) - not fully available
❓ FAQ
What is the package name for installation?
The package is now called ms-agent. For versions 0.8.0 or earlier, use modelscope-agent.
Does MS-Agent support MCP?
Yes, it supports agent chat with MCP (Model Context Protocol) for tool calling.
Can I generate videos with MS-Agent?
Yes, it supports short video generation of about 5 minutes via Singularity Cinema.
Is there a web interface?
Yes, a WebUI is available with React Router and FastAPI, providing SSE-streamed agent interaction.
📊 Repository
🤖 Overview, features, install steps and FAQ were generated from the project's README on Sep 4, 2026. Always check the original source before running commands.