Cooragent is an open-source AI agent collaboration community developed by the LeapLabTHU team. Its core goal is to help users quickly build agents, quickly build workflows, and quickly optimize workflows. Its core philosophy is that when building and optimizing agents becomes simple enough, the true AGI era will arrive.
Purpose
Cooragent is aimed at developers and users who need to create, manage, and collaborate with AI agents. It allows users to generate an agent by describing their needs in one sentence, and enables multiple agents to collaborate to complete complex tasks. Users can also publish created agents to the community to share with others.
Main Capabilities
Cooragent provides two task modes:
- Agent Factory mode: Simply describe the agent, and Cooragent will generate the agent according to the requirements. The system automatically analyzes user needs and, through memory and expansion, deeply understands the user, eliminating the need for complex Prompt design. The Planner selects appropriate tools, automatically polishes the Prompt, and gradually completes agent construction based on a deep understanding of user needs. After construction is complete, it can be used immediately, or its behavior and functions can continue to be edited and optimized.
- Agent Workflow mode: Simply describe the target task you want to complete, and Cooragent will automatically analyze the task requirements and select appropriate agents for collaboration. The Planner combines them according to each agent's expertise, plans the task steps and completion order, and then hands them to the task distribution node publish to publish the task. Each agent receives its own task and collaborates to complete it. Both modes can continuously evolve.
Cooragent also provides three workflow modes:
- Launch mode: Users only need to outline the desired task, and Cooragent automatically analyzes the requirements, selects appropriate agents, and builds a complete workflow. After the task is completed, the workflow is saved locally (usually located in store/workflow) for future reuse and modification. In the CLI tool, use the run-l command to start it.
- Polish mode: Provides fine-grained control. Users can manually optimize the workflow execution order, agent tool selection, LLM configuration, and related Prompts, or use natural language commands to let Cooragent make specific adjustments. For example, users can instruct: "Refine the stock analysis Agent: replace the browser tool with Tavily for faster information retrieval.” Cooragent automatically adjusts the agent's Prompt, tools, and other runtime parameters with the help of frameworks such as APE and Absolute-Zero-Reasoner. It is started with the run-o command in the CLI tool.
- Production mode: In this mode, Cooragent executes the optimized workflow with maximum efficiency, minimizing runtime overhead. The Supervisor module is used to ensure reliable output and handle exceptions. It is started with the run-p command in the CLI tool.
Best practices: Launch mode is used to automatically and quickly build a runnable workflow; Polish mode is used for fine-tuning and optimizing the workflow; Production mode is used for production scenarios and pursues stable operation.
In addition, Cooragent provides Cooragent Agent Studio, an online agent studio that can quickly generate agents capable of completing tasks in various scenarios, supports use anytime and anywhere, and allows managing and sharing created agents. It can be tried for free.
Usage
Cooragent provides a CLI developer tool to help developers quickly create, edit, and delete agents. The CLI design focuses on efficiency and ease of use, significantly reducing the tediousness of manual operations and allowing developers to focus more on the design and optimization of the agents themselves.
Installation methods support conda and venv. Note: When running the project CLI tool on Windows, in addition to the above steps, additional dependencies need to be installed. For details, please refer to Windows Platform Support. A .env file needs to be created in the project root directory and environment variables need to be configured.
Common CLI operation examples include:
- Enter the cooragent command tool interface
- Create a Xiaomi stock analysis agent with one command
- Optimize an agentic workflow (Polish an Agentic Workflow)
- Run an agentic workflow in Production mode
- Remove an agent
- Use a group of agents to collaborate on completing complex tasks
Integrating MCP Services
Cooragent supports integrating external services and tools through the Model Context Protocol (MCP) to enhance agents, similar to how desktop AI assistants such as Claude Desktop manage external functions. Configuration steps:
- Locate/createconfiguration file: Find or create config/mcp.json in the project root directory. You can run cd ./config and then cp mcp.json.example mcp.json.
- Add MCP services: Define MCP services in this JSON file. Each service has a unique key and configuration object. Example config/mcp.json: { "mcpServers": { "aws-kb-retrieval": { "command": "npx", "args": ["-y", "@modelcontextprotocol/server-aws-kb-retrieval"], "env": { "AWS_ACCESS_KEY_ID": "YOUR_ACCESS_KEY_HERE", "AWS_SECRET_ACCESS_KEY": "YOUR_SECRET_ACCESS_KEY_HERE", "AWS_REGION": "YOUR_AWS_REGION_HERE" } }, "AMAP": { "url": "https://mcp.amap.com/sse", "env": { "AMAP_MAPS_API_KEY": "AMAP_MAPS_API_KEY" } } } } After configuration is complete, Cooragent will register these MCP services as available tools. Agents can select and use these tools during task planning and execution, thereby enabling more complex functionality. Taking the configuration of AMAP mcpServers (Amap) as an example, related cases may then be able to run.
Applicable Scenarios
- Quickly generate agents for various scenarios
- Enable multiple agents to collaborate to complete complex tasks
- Build reusable and modifiable agent workflows
- Stably run optimized workflows in production environments
- Extend agent capabilities by integrating external services through MCP
Limitations
The materials do not provide specific technical limitations, pricing details, platform requirements, etc. Cooragent's specific availability, performance, and compatibility should be based on the official documentation and actual use. The project provides documentation and support, including Frequently Asked Questions (FAQ) and commercial support plans. Contributions of all forms are welcome, including fixing typos, improving documentation, or adding new features; see the contribution guidelines. The community provides a WeChat group, and the core contributors include Zheng Wang, Shenzhi Wang, Yue Wu, Shiji Song, and Gao Huang.
