✨ litemcp

A minimal, lightweight client designed to simplify SDK adoption into MCP.

litemcp enables rapid and intuitive integration of various AI SDKs (e.g., LangChain, Agent SDK) into your MCP projects, emphasizing simplicity, flexibility, and minimal dependencies.

🌟 Key Features

Simplicity: Streamlined interfaces ensure easy integration.
Flexibility: Quickly adopt diverse SDKs with minimal effort.
Lightweight: Designed with minimal dependencies to maximize clarity and performance.

🛠 Installation

Install via pip:

pip install litemcp

🚀 Quick Start

litemcp allows you to integrate tools from an MCP server into various LLM runtimes, including the OpenAI Agent SDK, LangChain, and direct OpenAI API calls.

Below are three examples showing how to use litemcp in different contexts:

✅ OpenAI Agent SDK Integration

async def main():
    async with MCPServerManager(sys.argv[1]) as server_manager:
        mcp_server_tools = await server_manager.agent_sdk_tools()

        agent = Agent(
            name="assistant",
            instructions="You are an AI assistant.",
            tools=mcp_server_tools,
        )

        result = await Runner.run(agent, "List all the kubernetes clusters")
        print(result.final_output)

if __name__ == "__main__":
    asyncio.run(main())

✅ LangChain Integration

async def main(config):
    chat = ChatOpenAI(model="gpt-3.5-turbo-0125")
    async with MCPServerManager(config) as server_manager:

        # bind tools
        tools: List[BaseTool] = await server_manager.langchain_tools()
        chat_with_tools = chat.bind_tools(tools, tool_choice="any")

        messages = [
            SystemMessage(content="You're a helpful assistant"),
            HumanMessage(content="List the dirs in the /Users"),
        ]
        tool_calls = chat_with_tools.invoke(messages).tool_calls

        # invoke the tool_call
        tool_map = {tool.name: tool for tool in tools}
        for tool_call in tool_calls:
            selected_tool = tool_map[tool_call["name"].lower()]
            tool_output = await selected_tool.ainvoke(tool_call["args"])
            print(tool_output)

✅ Direct OpenAI API Integration

async def main(config):
    client = OpenAI()

    async with MCPServerManager(config) as server_manager:
        schemas = await server_manager.schemas()

        completion = client.chat.completions.create(
            model="gpt-4o",
            messages=[{"role": "user", "content": "List the dirs in the /Users"}],
            tools=schemas,
        )

        print(completion.choices[0].message.tool_calls)

        # Execute the selected tool
        tool_call = completion.choices[0].message.tool_calls[0]
        result = await server_manager.tool_call(
            tool_call.function.name, tool_call.function.arguments
        )
        print(result.content[0].text)

🔐 Tool Call Validator(Optional)

You can add a custom validation function to control MCP tool calls. This helps prevent server tools from directly accessing your system without permission—such as integrating a human-in-the-loop step.

1. Define the Validator

def applier_validator(func_args) -> Optional[str]:
    """
    Return:
    - None: allow the tool call
    - str : block the tool call and return message instead
    """
    user_input = console.input(
        f"  🛠  Cluster - [yellow]{cluster}[/yellow] ⎈ Proceed with this YAML? (yes/no): "
    ).strip().lower()

    if user_input in {"yes", "y"}:
        return None
    if user_input in {"no", "n"}:
        console.print("[red]Exiting process.[/red]")
        sys.exit(0)
    return user_input

2. Register the Validator with MCP Server

async with MCPServerManager(sys.argv[1]) as server_manager:
    server_manager.register_validator("yaml_applier", applier_validator)

    mcp_server_tools = await server_manager.agent_sdk_tools()

    engineer = Agent(...)

📖 MCP Configuration Schema

Configure your MCP environment with optional server enabling and tool exclusion:

{
  "mcpServers": {
    "fetch": {
      "command": "uvx",
      "args": ["mcp-server-fetch"]
    },
    "youtube": {
      "command": "npx",
      "args": ["-y", "github:anaisbetts/mcp-youtube"],
      "exclude_tools": ["..."]
    },
    "mcp-server-commands": {
      "command": "npx",
      "args": ["mcp-server-commands"],
      "requires_confirmation": [
        "run_command",
        "run_script"
      ],
      "enabled": false
    },
     "multicluster-mcp-server": {
      "command": "node",
      "args": [".../multicluster-mcp-server/build/index.js"],
      "enabled": false
    }
  }
}

Use "enabled": true/false to activate or deactivate servers.
Use "exclude_tools" to omit unnecessary tools from the current MCP server.

📖 Documentation

Detailed documentation coming soon!

📢 Contributing

Contributions and suggestions are welcome! Please open an issue or submit a pull request.

📜 License

liteMCP is available under the MIT License.

litemcp