MCP Client

This project is an implementation of an MCP (Model Context Protocol) client for OpenAI Agents SDK.

Features

• Dynamic and Flexible Configuration: Effortlessly add or modify MCP servers through a simple configuration file. Explore a curated list of powerful MCP servers here and orchestrate them seamlessly to suit your needs.
• Supports agent orchestration, allowing multiple agents to collaborate and delegate tasks.
• Easy integration with OpenAI Agents SDK.
• Scalable design to handle complex workflows with multiple agents.
• Customizable agent instructions and behavior.

Requirements

• Python 3.8 or higher
• OpenAI Agents SDK (pip install openai-agents)

Usage

Make configuration file (config.json)

{
  "agents": [
    {
      "name": "agent1",
      "model": "model_name",
      "instructions": "Agent 1 instructions",
      "cache_tools_list": true,
      "assistants": ["agent2"],
      "mcp_servers": [
        {
          "command": "path/to/command",
          "args": ["--arg1", "value1"]
        }
      ]
    },
    {
      "name": "agent2",
      "model": "model_name",
      "instructions": "Agent 2 instructions",
      "cache_tools_list": false,
      "assistants": [],
      "mcp_servers": [
        {
          "command": "path/to/command",
          "args": ["--arg2", "value2"]
        }
      ]
    }
  ],
  "starting_agent": "agent1"
}

Explanation of Fields

• agents: A list of agent configurations.
  • name: The unique name of the agent.
  • model: The model associated with the agent.
  • instructions: Instructions or description for the agent.
  • cache_tools_list: A boolean indicating whether to cache the tools list.
  • assistants: A list of other agents that this agent can hand off tasks to.
  • mcp_servers: A list of MCP server configurations.
    • command: The command to start the MCP server.
    • args: A list of arguments for the command.
• starting_agent: The name of the agent to start the workflow.

Notes

• Ensure that all agents listed in the assistants field exist in the agents list.
• Warning: Circular dependencies between agents may lead to infinite loops during execution. Ensure that the configuration avoids such scenarios.

Run the client with:

export OPENAI_API_KEY=<Your API KEY>
python main.py <config_file>

Use exit, quit, or q to terminate the program during execution.