agents
MCP Multi-Framework Agent Integration Demo and Benchmarking Platform
Install & Use
Copy this prompt and send it to your AI assistant (Claude / Cursor / TRAE / Codex / WorkBuddy etc.) to auto-install:
Help me install this AI Skill: agents. It is used for: MCP Multi-Framework Agent Integration Demo and Benchmarking Platform Full Skill content: https://321skill.com/skills/agents-agents-mcp-usage/raw/index.md Read that page and install it.
The prompt includes a link to the full Skill content. You can also view the full content.
This Skill addresses the lack of unified examples and benchmarking tools for developers integrating MCP (Model Context Protocol) servers into different LLM Agent frameworks. In practice, teams often need to compare MCP integration approaches across frameworks like Google ADK, LangGraph, OpenAI Agents, and Pydantic-AI Agents, but they lack reusable reference code and performance comparison data.
Usage is straightforward: simply clone the repository, configure environment variables (e.g., GEMINI_API_KEY, OPENAI_API_KEY), and run a single example script. For instance, execute uv run agents_mcp_usage/basic_mcp/basic_mcp_use/pydantic_mcp.py to experience a Pydantic-AI Agent invoking MCP tools. You can also run make benchmark to launch Dockerized benchmark tests and use Logfire tracing visualization to observe the Agent's call chain.
It is ideal for agent developers, full-stack engineers, and AI researchers who need to evaluate and select LLM Agent frameworks. Teams already familiar with the MCP protocol but unsure how to implement it across different frameworks can use this platform to quickly compare integration complexity, execution efficiency, and extensibility.
It is recommended to run the platform's benchmark tests to compare framework performance before building a new Agent system. Note that this project is primarily for demonstration and benchmarking purposes, not for production use. For actual deployment, adjust MCP server configurations based on business needs and ensure secure API key management.
Key Features
Unlike examples for MCP integration with a single framework, this repository simultaneously supports four mainstream frameworks—Google ADK, LangGraph, OpenAI Agents, and Pydantic-AI Agents—and includes a unified evaluation suite with Logfire tracing visualization, enabling developers to directly compare call chains and performance metrics across different frameworks.
Limitations
Requires a Python 3.10+ environment and the uv package manager. Depends on Gemini or OpenAI API keys. Currently only supports basic MCP server examples and does not cover advanced scenarios like streaming or custom error handling.
FAQ
How do I add a new MCP server?
Refer to `mcp_servers/example_server.py` to create a custom MCP server, then register it in your Agent script via `mcp.use_server()` or the corresponding method for your framework.
Which LLM models are supported?
Each framework defaults to using Gemini or OpenAI models. You can switch to other models via environment variables or code parameters in the scripts, provided the framework supports them.
Installation guide for AI assistants
If your AI coding assistant (Claude Code, Cursor, TRAE etc.) can see this page, send it this message to auto-install:
Visit https://321skill.com/skills/agents-agents-mcp-usage/raw/index.md to read the original Skill definition (Markdown format) for agents, and install it according to the instructions.
Raw Markdown URL for AI: /skills/agents-agents-mcp-usage/raw/index.md