mcp-demo

# Examples with Model Context Protocol (MCP) This repository provides examples demonstrating the integration of Gemini models with custom tools, specifically highlighting their interaction within a **Model Context Protocol (MCP)** environment. It also showcases different methods for exposing and consuming these functionalities. **Understanding Model Context Protocol (MCP):** In this context, the Model Context Protocol (MCP) likely defines a standardized way for language models (like Gemini) to interact with external tools. This protocol might specify: - How tools are described and discovered by the model. - The format for invoking tools, including necessary parameters. - The structure of the responses expected from tools. - Mechanisms for managing the context and state during tool interactions. These examples aim to illustrate how to build and utilize tools that seamlessly integrate with an MCP-based system. ## Examples This repository contains the following examples: ### 1. FastAPI Server Exposing Tools for MCP This example demonstrates how to expose your FastAPI endpoints as **Model Context Protocol (MCP)** tools, with Auth. This allows Gemini models, operating within an MCP framework, to discover and utilize these custom functionalities. - **File:** `fast_api_mcp_demo.py` - **References:** - [Fast-API MCP](https://github.com/tadata-org/fastapi_mcp) ### 2. Gemini Model Consuming MCP Tools This example showcases how a Gemini model, operating within an **MCP** environment, can consume the tools exposed by an MCP-compliant server. This demonstrates the complete cycle of a Gemini model discovering, invoking, and utilizing external tools through the defined protocol. - **File:** `gemini_mcp_demo.py` - **Key Features:** - Demonstrates the process of an MCP-enabled Gemini model discovering available tools. - Provides code snippets showing how to format tool invocation requests according to the MCP specification. - Likely includes examples of processing the structured responses from the executed MCP tools and integrating them into the model's reasoning. - **References:** - [Using MCP with Gemini](https://ai.google.dev/gemini-api/docs/function-calling?example=meeting#use_model_context_protocol_mcp) ### 3. SDK Server with Server-Sent Events (SSE) for MCP Interaction This example illustrates creating a server using an SDK and exposing functionalities via Server-Sent Events (SSE) in the context of **MCP**. While SSE is a transport mechanism, this example likely shows how it can be used to facilitate communication or stream information relevant to the MCP workflow. - **File:** `mcp_server_demo.py` - **Key Features:** - Shows the server-side implementation using the specified SDK, focusing on its role within an MCP architecture. - Illustrates how to configure and send events using SSE that might be relevant to the MCP interaction lifecycle. - May include examples of how an MCP agent or the Gemini model itself could consume this SSE stream for real-time updates or information related to tool execution. ### Prerequisites Before running the examples, ensure you have the following installed: - Python >= 3.10 - Pip (Python package installer) - Any specific libraries required by each example (listed in their respective requirements files or documentation). - Access to the Gemini models and the necessary infrastructure to interact within an **MCP** environment. This might involve specific SDKs or API endpoints related to your MCP implementation. - (For the FastAPI example) A running MCP server or the necessary components to simulate its behavior and interaction with the exposed tools. ## Contributing Contributions are welcome! If you find bugs or have suggestions for improving the application, please open an issue or submit a pull request. Made with ❤ by [jggomez](https://devhack.co). [](https://twitter.com/jggomezt) [](https://www.linkedin.com/in/jggomezt/) [](https://medium.com/@jggomezt) ## License Copyright 2025 Juan Guillermo Gómez