adk -gemma3

This MCP server implements an YouTube search assistant built with Google's ADK, MCP, and Gemma 3 running via Ollama. The agent uses MCP to interact with a YouTube search tool, call the SERP API for results, and then have Gemma 3 (via Ollama) format and present the results in a readable, bullet-point format. You can run the agent directly with Python and access a web UI for debugging if you enable the ADK web interface, or interact via the command line to issue natural language search queries such as asking for YouTube videos on a topic, a tutorial, or a specific event. The MCP tooling ensures standardized input/output between the agent and the YouTube search tool, so the model can chain calls and present well-structured results.

Prerequisites:\n- Python 3.9+\n- A SERP API key for YouTube search\n- Ollama installed with Gemma 3 model (for Gemma 3 via Ollama)\n\nStep-by-step:\n1) Clone the repository and enter the project directory:\nbash\ngit clone https://github.com/arjunprabhulal/adk-mcp-gemma3.git\ncd adk-mcp-gemma3\n``\n\n2) Create and activate a virtual environment:\nbash\npython -m venv .venv\n# macOS/Linux\nsource .venv/bin/activate\n# Windows\n.venv\Scripts\activate\n\n\n3) Install dependencies:\nbash\npip install -r requirements.txt\n\n\n4) Set up your SERP API key:\nbash\n# In the project root, create .env or set env variable in your shell\nSERP_API_KEY=your_serp_api_key_here\n\n\n5) Pull the Gemma 3 model for Ollama:\nbash\nollama pull gemma3:12b\n\n\n6) Run the MCP server (as per the README, the server is started with the MCP entry point):\nbash\n# Start the MCP-enabled search agent using the module entry point\npython -m search\n```\n\nOptional: If you want to use the web UI for debugging, navigate to the parent directory and run the ADK web interface as described in the README.\n

Tips and caveats:\n- Ensure Ollama is running and Gemma 3 is accessible before starting the MCP server.\n- Set SERP_API_KEY in a secure way; the key is required for YouTube search via SERP API.\n- If you modify environment variables, restart the MCP server to pick up changes.\n- The MCP tool is used to standardize tool invocation; you should see structured tool calls in the UI or logs when queries trigger YouTube searches.\n- If you encounter model- or tool-call formatting issues, verify that Gemma 3 is correctly mounted via Ollama and that the LiteLlm configurations are compatible with your ADK version.\n- The current setup assumes a Python-based ADK/MCP pipeline; adjust the command path if you restructure the project layout.