AI Agents Overview¶
The Agentic Networking Quickstart demonstrates the power of declarative authorization policies using the Multi-Server MCP Client. To help users get started, the project provides three different AI agents, demonstrating various ways to build and integrate these capabilities.
ADK Agent¶
The ADK Agent is a lightweight agent built with Google's Agent Development Kit (ADK). It uses ADK's McpToolset to connect to multiple MCP servers and LiteLLM to interact with a variety of language models, executing tools exposed by the connected MCP servers.
LangChain Agent¶
The LangChain Agent demonstrates how to integrate with the Multi-Server MCP Client using the popular Langchain framework. The agent is built using langchain, litellm, and streamlit to provide a user-friendly chat interface. The LiteLLM integration gives you the flexibility to easily connect to a variety of language models, including Hugging Face models.
OpenAI Agent¶
The OpenAI Agent demonstrates how to build an agent using the official OpenAI SDK. It uses the SDK to communicate with a language model and dynamically loads tools from connected MCP servers. Built using Gradio to provide an intuitive web chat interface, it connects to any OpenAI-compatible endpoint—meaning you can use it with OpenAI's models or open-source models hosted on platforms like Hugging Face.