AIAgent-MCP

# AIAgent-MCP: Agentic AI App with MCP and OpenAPI Integration ## Additional details about this POC is provided in this [blog](https://medium.com/beyond-the-buzz-highlighting-the-impact-of-ai-in/the-lazy-devs-guide-to-building-a-local-ai-agent-that-actually-works-30e7f3aee140). ## This project demonstrates a lightweight agentic AI system powered by a quantized `gemma3:12b` model running on Ollama, with tool integration via the MCP server. The goal is to test tool-augmented reasoning like fetching current time, summarizing articles, or retrieving the latest AI news. --- ## 🚀 Getting Started ### 1. Clone the Git Repository ```bash git clone https://github.com/AhilanPonnusamy/AIAgent-MCP.git cd AIAgent-MCP ``` ### 2. Create and Activate a Virtual Environment Use Python 3.11 and the following naming convention to avoid updating configs manually. ```bash python3.11 -m venv mcplatest-venv source mcplatest-venv/bin/activate ``` ### 3. Start the Ollama Server and Model Run the Gemma 3 12B model locally using Ollama: ```bash ollama run gemma3:12b ``` ### 4. Start the MCPO Tool Server ```bash uvx mcpo --config config.json --port 8001 ``` Access Tool Endpoint: http://localhost:8001/ainews/docs ### 5. Start the Agent (Client) In a new terminal (inside virtual environment): ```bash source mcplatest-venv/bin/activate uvicorn agentic_client:app --reload --port 8000 ``` ### 6. Start the Streamlit UI In another terminal (also inside virtual environment): ```bash source mcplatest-venv/bin/activate streamlit run app-ui.py ``` ## Example Prompts to Try Start chatting from the Streamlit UI! #### "What is the distance to the moon?" #### "What is the current time?" → Should trigger the time tool. #### "Can you get latest AI news?" → Should trigger ainews tool and return top 10 articles. #### "Can you summarize this article for me: Efficient Fine-Tuning of Language Models with Low-Rank Adapters - https://arxiv.org/abs/2405.16746?" → Should invoke fetch tool and return a nice summary. >[!WARNING] > Due to the nature of low precision quantized models, behavior may sometimes be inconsistent. When in doubt: restart servers and LLM (ollama stop) > things should be back to normal! 🎉 Have Fun! This project is your sandbox for building powerful local agentic systems with reasoning and tool-usage capabilities. Extend, explore, and enjoy! 😄