OpenAI_Agent_SDK_plus_MCP

# Model Context Protocol (MCP) + OpenAI Agent SDK This repository demonstrates how to leverage the Model Context Protocol (MCP) within the OpenAI Agent SDK to build context-aware applications. It includes three example implementations: - **Local Agent**: Connects to a local MCP server using Standard Input and Output (STDIO). - **Remote Agent**: Connects to a remote MCP server using Server-Sent Events (SSE). - **Voice Agent**: Integrates with an MCP server to enable voice-based interactions. The examples are organized into the following directories: - `mcp_server_sse`: Contains the implementation for the SSE-based remote agent. - `mcp_server_stdio`: Contains the implementation for the STDIO-based local agent. - `mcp_voice_agent`: Contains the implementation for the voice-enabled agent. ## Local Agent To run the `local agent` example, you only need to install the OpenAI Agent SDK: ```bash pip install openai-agents ``` You don't need to install the MCP SDK separately, as it is already included as a dependency of the OpenAI Agent SDK. Once the installation is complete, navigate to the `mcp_server_stdio` directory and run the `main.py` script: ```bash python main.py ``` This would launch a `REPL` that would let you use the [Filesystem MCP server](https://github.com/modelcontextprotocol/servers/tree/main/src/filesystem) to interact with the `sample_folder` ## Remote Agent The installation for the remote agent example is the same. Although this example depends on `uvicorn` and `starlette` but these are already dependencies of the MCP SDK. To run this example first run the MCP server which is in the `server.py` file. ```bash python server.py ``` Then run the agent as well which is in the `main.py` file. ```bash python main.py ``` Again this launches an `REPL` but instead of connecting to an MCP server via standard Input and Output it connects to the server running on local host via SSE. ## Voice Agent The installation process for the voice Agent is a bit different you need to install the OpenAI Agent SDK with voice dependencies. ```bash pip install 'openai-agents[voice]' ``` Then install sounddevice: ```bash pip install sounddevice ``` With this you can now run the the `main.py` python script. ```bash python main.py ``` This will launch a `REPL` with a voice-based interface, allowing you to interact with the MCP server using voice commands instead of text.