MCP Server Markup Language (MCPML)

A Python framework for building Model Context Protocol (MCP) servers with CLI and OpenAI Agent support.

Features

• 🚀 MCP Server Framework: Build MCP-compliant servers in Python
• 🔧 CLI Tools: All server capabilities exposed as CLI commands (to be consumed by humans or scripts rather than MCP clients)
• 🤖 OpenAI Agent SDK Support: Implement tools as OpenAI agents Or as simple python functions
• 🔄 Agent-to-MCP Integration: Agents can consume MCP services via config
• 🛠️ Extensible Architecture: Easily add custom tools and services
• 🔌 Dynamic Loading: Support for custom agent types and tool implementations from the execution directory
• 📦 Structured Output: Support for structured output using Pydantic models

Installation

pip install git+https://github.com/a5c-ai/mcpml#egg=mcpml

.env

OPENAI_API_KEY=your_openai_api_key

or

AZURE_OPENAI_ENDPOINT=https://your-azure-openai-endpoint.openai.azure.com
AZURE_OPENAI_API_KEY=your_azure_openai_api_key
OPENAI_API_VERSION=api_version

Usage

mcpml --help

mcpml run

mcpml.yaml is the default config file for the MCPML server.

mcpml run -c https://github.com/a5c-ai/some-mcpml-server

mcpml -c mcpml.yaml tools some-tool run --arg1 value1 --arg2 value2

using uvx:

uvx --from git+https://github.com/a5c-ai/mcpml#egg=mcpml mcpml -c mcpml.yaml tools list

uvx --from git+https://github.com/a5c-ai/mcpml#egg=mcpml mcpmp run --transport=sse

License

MIT