mcp-kit 0.2.2

MCP tooling for developing and optimizing multi-agent AI systems.

MCP Kit Python

MCP tooling for developing and optimizing multi-agent AI systems

A comprehensive toolkit for working with the Model Context Protocol (MCP), providing seamless integration between AI agents and various data sources, APIs, and services. Whether you're building, testing, or deploying multi-agent systems, MCP Kit simplifies the complexity of tool orchestration and provides powerful mocking capabilities for development.

Features

Flexible Target System

• MCP Servers: Connect to existing MCP servers (hosted or from specifications)
• OpenAPI Integration: Automatically convert REST APIs to MCP tools using OpenAPI/Swagger specs
• Mock Responses: Generate realistic test data using LLM or random generators
• Multiplexing: Combine multiple targets into a unified interface

Framework Adapters

• OpenAI Agents SDK: Native integration with OpenAI's agent framework
• LangGraph: Seamless tool integration for LangGraph workflows
• Generic Client Sessions: Direct MCP protocol communication
• Official MCP Server: Standard MCP server wrapper

Configuration-Driven Architecture

• YAML/JSON Configuration: Declarative setup for complex workflows
• Factory Pattern: Clean, testable component creation
• Environment Variables: Secure credential management

Advanced Response Generation

• LLM-Powered Mocking: Generate contextually appropriate responses using LLMs
• Random Data Generation: Create test data for development and testing
• Custom Generators: Implement your own response generation logic

---

Quick Start

Installation

uv add mcp-kit

Basic Usage

First you write the Proxy config:

# proxy_config.yaml
""" A mocked REST API target given the OpenAPI spec using LLM-generated responses
"""
target:
  type: mocked
  base_target:
    type: oas
    name: base-oas-server
    spec_url: https://petstore3.swagger.io/api/v3/openapi.json
  tool_response_generator:
    type: llm
    model: openai/gpt-4.1-nano

Don't forget to setup the LLM API KEY:

# .env
OPENAI_API_KEY="your_openai_key"

Then we can use it as any other MCP:

# main.py
from mcp_kit import ProxyMCP


async def main():
    # Create proxy from configuration
    proxy = ProxyMCP.from_config("proxy_config.yaml")

    # Use with MCP client session adapter
    async with proxy.client_session_adapter() as session:
        tools = await session.list_tools()
        result = await session.call_tool("get_pet", {"pet_id": "777"})
        print(result)


if __name__ == "__main__":
    import asyncio

    asyncio.run(main())

---

Core Concepts

Targets

Targets are the core abstraction in MCP Kit, representing different types of tool providers:

MCP Target

Connect to existing MCP servers:

target:
  type: mcp
  name: my-mcp-server
  url: http://localhost:8080/mcp
  headers:
    Authorization: Bearer token123

OpenAPI Target

Convert REST APIs to MCP tools:

target:
  type: oas
  name: petstore-api
  spec_url: https://petstore3.swagger.io/api/v3/openapi.json

Mocked Target

Generate fake responses for testing:

target:
  type: mocked
  base_target:
    type: mcp
    name: test-server
    url: http://localhost:9000/mcp
  tool_response_generator:
    type: llm
    model: openai/gpt-4.1-nano

Multiplex Target

Combine multiple targets:

target:
  type: multiplex
  name: combined-services
  targets:
    - type: mcp
      name: weather-service
      url: http://localhost:8080/mcp
    - type: oas
      name: petstore
      spec_url: https://petstore3.swagger.io/api/v3/openapi.json

Adapters

Adapters provide framework-specific interfaces for your targets:

• Client Session Adapter: Direct MCP protocol communication
• OpenAI Agents Adapter: Integration with OpenAI's agent framework
• LangGraph Adapter: Tools for LangGraph workflows
• Official MCP Server: Standard MCP server wrapper

Response Generators

Response Generators create mock responses for testing and development:

• LLM Generator: Uses language models to generate contextually appropriate responses
• Random Generator: Creates random test data
• Custom Generators: Implement your own logic

---

Examples

OpenAI Agents SDK Integration

from mcp_kit import ProxyMCP
from agents import Agent, Runner, trace
import asyncio

async def openai_example():
    proxy = ProxyMCP.from_config("proxy_config.yaml")

    async with proxy.openai_agents_mcp_server() as mcp_server:
        # Use with OpenAI Agents SDK
        agent = Agent(
            name="research_agent",
            instructions="You are a research assistant with access to various tools.",
            model="gpt-4.1-nano",
            mcp_servers=[mcp_server]
        )

        response = await Runner.run(
            agent,
            "What's the weather like in San Francisco?"
        )
        print(response.final_output)

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

LangGraph Workflow Integration

from mcp_kit import ProxyMCP
from langgraph.prebuilt import create_react_agent
import asyncio

async def langgraph_example():
    proxy = ProxyMCP.from_config("proxy_config.yaml")

    # Get LangChain-compatible tools
    client = proxy.langgraph_multi_server_mcp_client()

    async with client.session("your_server_name") as _:
        # Get the MCP tools as LangChain tools
        tools = await client.get_tools(server_name="your_server_name")

        # Create ReAct agent
        agent = create_react_agent(model="google_genai:gemini-2.0-flash", tools=tools)

        # Run workflow
        response = await agent.ainvoke({
            "messages": [{"role": "user", "content": "Analyze Q1 expenses"}]
        })

        # Extract result
        final_message = response["messages"][-1]
        print(final_message.content)

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

Testing with Mocked Responses

from mcp_kit import ProxyMCP
import asyncio

async def testing_example():
    # Configuration with LLM-powered mocking
    proxy = ProxyMCP.from_config("proxy_config.yaml")

    async with proxy.client_session_adapter() as session:
        # These calls will return realistic mock data
        tools = await session.list_tools()
        expenses = await session.call_tool("get_expenses", {"period": "Q1"})
        revenues = await session.call_tool("get_revenues", {"period": "Q1"})

        print(f"Available tools: {[tool.name for tool in tools.tools]}")
        print(f"Mock expenses: {expenses}")
        print(f"Mock revenues: {revenues}")

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

Configuration Examples

Development with Mocking

# dev_proxy_config.yaml
target:
  type: mocked
  base_target:
    type: oas
    name: accounting-api
    spec_url: https://api.company.com/accounting/openapi.json
  tool_response_generator:
    type: llm
    model: openai/gpt-4.1-nano

Production MCP Server

# prod_proxy_config.yaml
target:
  type: mcp
  name: production-accounting
  url: https://mcp.company.com/accounting
  headers:
    Authorization: Bearer ${PROD_API_KEY}
    X-Client-Version: "1.0.0"

Multi-Service Architecture

# multi_proxy_config.yaml
target:
  type: multiplex
  name: enterprise-tools
  targets:
    - type: mcp
      name: crm-service
      url: https://mcp.company.com/crm
    - type: oas
      name: analytics-api
      spec_url: https://api.company.com/analytics/openapi.json
    - type: mocked
      base_target:
        type: mcp
        name: experimental-service
        url: https://beta.company.com/mcp
      tool_response_generator:
        type: random

---

Advanced Configuration

Environment Variables

# .env file
OPENAI_API_KEY=your-openai-key
WEATHER_API_KEY=your-weather-key

Custom Response Generators

from mcp_kit.generators import ToolResponseGenerator
from mcp_kit import ProxyMCP

class CustomGenerator(ToolResponseGenerator):
    async def generate(self, target_name: str, tool: Tool, arguments: dict[str, Any] | None = None) -> list[Content]:
        # Your custom logic here
        return [TextContent(type="text", text=f"Custom response for {tool.name} on {target_name}")]

# Use in configuration
proxy = ProxyMCP(
    target=MockedTarget(
        base_target=McpTarget("test", "http://localhost:8080"),
        mock_config=MockConfig(tool_response_generator=CustomGenerator())
    )
)

---

Project Structure

mcp-kit-python/
├── src/mcp_kit/
│   ├── adapters/          # Framework adapters
│   │   ├── client_session.py
│   │   ├── openai.py
│   │   └── langgraph.py
│   ├── generators/        # Response generators
│   │   ├── llm.py
│   │   └── random.py
│   ├── targets/          # Target implementations
│   │   ├── mcp.py
│   │   ├── oas.py
│   │   ├── mocked.py
│   │   └── multiplex.py
│   ├── factory.py        # Factory pattern implementation
│   └── proxy.py          # Main ProxyMCP class
├── examples/             # Usage examples
│   ├── openai_agents_sdk/
│   ├── langgraph/
│   ├── mcp_client_session/
│   └── proxy_configs/
└── tests/               # Test suite

---

Contributing

We welcome contributions! Please see our Contributing Guide for details.

Development Setup

git clone https://github.com/agentiqs/mcp-kit-python.git
cd mcp-kit-python
uv sync --dev
pre-commit install

Running Tests

uv run pytest tests/ -v

---

Documentation

• Installation Guide
• Configuration Reference
• Framework Adapters
• API Reference
• Examples

---

License

This project is licensed under the Apache License 2.0 - see the LICENSE file for details.

---

Support

• GitHub Issues
• Documentation
• Examples

---

Built with ❤️ by Agentiqs