gemini_mcp

🤖 Gemini MCP Agent – Modular Tool-Using AI in Streamlit

This project is a LangChain-style, Model Context Protocol (MCP) compliant AI agent built on Google Gemini Pro, supporting dynamic tools, code execution, file reading, and more — all inside a clean Streamlit app.

---

Screen Shots

🚀 Features

✅ Gemini Pro API integration
✅ MCP-style prompt + tool schema injection
✅ Agent runtime with plan → act → observe loop
✅ Tool output formatted in Markdown/code blocks
✅ Dynamic tool discovery (just drop in new files)
✅ Built-in tools:

• 📄 readfile → PDF/DOCX parser
• 🌐 scrape → web page content extractor
• 📚 wikipedia → 3-sentence topic summary
• 🧮 code → secure Python code execution

---

🧠 Architecture Overview

User Prompt → Streamlit UI → MCP Agent
           ↘︎ Memory ↘︎ Tool Schemas ↘︎
        → Gemini Pro API → JSON response → Tool call / Final Answer

Layer	Implementation
LLM API	Gemini via google.generativeai
Prompt Pipeline	MCP format with tool schema JSON
Agent Runtime	Custom loop: plan → act → observe
Orchestration	Modular Streamlit + dynamic tools

---

🛠️ Tools Directory

All tools inherit from BaseTool and live in tools/.

Tool schema auto-discovery is handled dynamically — no need to manually register.

To add a new tool:

# tools/my_tool.py

from tools.base_tool import BaseTool

class MyTool(BaseTool):
    name = "mytool"
    description = "Do something useful"
    parameters = {
        "type": "object",
        "properties": {"input": {"type": "string", "description": "Some input"}},
        "required": ["input"]
    }

    def run(self, **kwargs) -> str:
        return f"You said: {kwargs['input']}"

---

🖥️ How to Run

1. Clone the repo and install requirements

git clone https://github.com/yourname/gemini-mcp-agent.git
cd gemini-mcp_agent
pip install -r requirements.txt

2. Set your Gemini API key

In app.py, replace:

API_KEY = "YOUR_API_KEY_HERE"

with your actual key from: https://makersuite.google.com/app/apikey

Or load it from .env.

---

3. Run the app

streamlit run app.py

---

💬 Example Prompts

• Wikipedia: Isaac Newton
• Scrape https://www.example.com
• Upload a file, then say: Summarize this document
• What is 32 * 17?
• Use code to compute factorial of 5

---

🧩 Coming Next?

• .eml email file parsing
• Vector memory / RAG from file chunks
• FastAPI deployment backend
• LangGraph-style multi-agent workflows

---

📄 License

MIT — free to use, modify, and extend.

---

🙌 Acknowledgments

• Google Generative AI SDK
• Streamlit
• LangChain MCP Protocol

---

Built with ❤️ to explore agentic LLMs and modular AI design.