mcp_kit

MCP Kit: A comprehensive toolkit for working with Model Context Protocol (MCP) servers.

This package provides a flexible framework for mocking, testing and optimizing agent interactions with the world by converting any target to the widely adopted Model Context Protocol and proxying tool calls where we can choose to mock or pass through. It includes support for various target interaction types, mock response generators, and adapter patterns to integrate with different AI frameworks and tools.

Key Components:

• Targets: Different types of interaction endpoints (MCP servers (hosted or spec), REST APIs (hosted or spec) and in the near future file systems, computer use, code interpreter, etc.)
• Adapters: Integration patterns for various frameworks (OpenAI Agents, LangGraph, Official MCP Client and Server)
• Generators: Response generation strategies for testing and mocking (random text or LLM-based responses)
• Proxy: Central interface providing multiple adapter patterns for a single target

Example:

from mcp_kit import ProxyMCP

# Create a proxy from configuration that contains a target definition and a mock strategy
proxy = ProxyMCP.from_config("proxy_config.yaml")

# Use as MCP Client Session
async with proxy.client_session_adapter() as session:
    tools = await session.list_tools()

    # Add tools to an LLM and catch tool calls
    ...

    # Execute a tool call
    await session.call_tool(
        name="example_tool",
        arguments={"param1": "value1", "param2": "value2"},
    )

# Use as OpenAI Agent
async with proxy.openai_agents_mcp_server() as mcp_server:
    agent = Agent(
        name="example_agent",
        instructions="This agent can call tools.",
        tools=tools,
        mcp_servers=[mcp_server],
    )

    response = await Runner.run(agent, "some input")

#  Use as LangGraph Multi-Server MCP Client
async with proxy.langgraph_multi_server_mcp_client() as client:
    # Get the MCP tools as LangChain tools
    tools = await client.get_tools(server_name="odoo_mcp_mocked")

    # Create the ReAct agent with the tools
    agent = create_react_agent(model="provider:model", tools=tools)

    # Run the agent
    response = await agent.ainvoke(
        {"messages": [{"role": "user", "content": "some input"}]},
    )