twittermcp

Gemini Tool-Aware CLI Chatbot with Model Context Protocol (MCP) This project is a Node.js-based Command Line Interface (CLI) chatbot that integrates Google’s Gemini AI with tool calling capability via the Model Context Protocol (MCP). It enables AI-assisted function calling, dynamic tool execution, and a conversational interface. This is ideal for developers looking to build AI agents that interact with APIs, tools, or services in real-time.

🧠 Project Overview This CLI chatbot uses:

Google Gemini (Generative AI) to process and generate conversational responses.

MCP (Model Context Protocol) to register and call external tools.

SSE (Server-Sent Events) for real-time tool interactions.

A simple interactive CLI interface powered by Node.js and readline.

The AI receives context, invokes tools as needed, and returns enhanced responses — all in a chat loop.

📁 Project Structure bash Copy Edit client/ │ ├── index.js # Main entry point ├── package.json # Project metadata and dependencies ├── .env # Environment variables (Gemini API key) └── README.md # Documentation (this file) 🚀 Features ✅ Text-based CLI interface

✅ Gemini 2.0 Flash model integration

✅ Supports function/tool calling using Gemini's functionDeclarations

✅ Dynamic tool listing via MCP

✅ Tool result injection into chat history

✅ Auto-reinvocation of Gemini after tool execution

✅ Clean error handling (including Gemini 503 responses)

✅ Fully async flow

🔧 Technologies Used Tool / Library Purpose @google/genai Connects to Google Gemini API @modelcontextprotocol/sdk Interacts with tool server via SSE dotenv Manages API keys securely readline/promises Asynchronous CLI input handling Node.js (ESM) Runtime & ecosystem

⚙️ Prerequisites Before running the project, ensure you have the following installed:

Node.js v18+

A valid Google Gemini API Key

An MCP-compatible tool server running at http://localhost:3001/sse

📦 Installation Clone the repository:

bash Copy Edit git clone https://github.com/your-username/gemini-mcp-chatbot.git cd gemini-mcp-chatbot Install dependencies:

bash Copy Edit npm install Create .env file:

env Copy Edit GEMINI_API_KEY=your_google_gemini_api_key_here 🧠 How It Works — Step-by-Step

1. Start and Connect When the app starts, it connects to the MCP server using Server-Sent Events (SSEClientTransport).

It then dynamically retrieves all registered tools via listTools() and registers their schemas with Gemini.

2. Read User Input It uses readline.promises to prompt the user in the terminal.

Input is added to the chatHistory.

3. Call Gemini Model The generateContent method from Google Gemini is called.

It receives the chat history and list of tools (functionDeclarations) as context.

4. Function Call Handling If Gemini chooses to invoke a tool, it returns a functionCall object.

The tool name and arguments are extracted and sent to the MCP server via mcpClient.callTool.

5. Insert Tool Result into Chat The result returned from the tool is inserted back into the conversation as if the user provided it.

The updated history is passed to Gemini again for a follow-up response.

6. Show Response If there was no tool invocation, the response is simply printed to the terminal.

The loop continues, waiting for the next user input.

🔍 Example Tool Call Let’s say the tool server has a tool named getWeather:

Tool Schema: json Copy Edit { "name": "getWeather", "description": "Returns weather based on city", "parameters": { "type": "object", "properties": { "city": { "type": "string" } }, "required": ["city"] } } Conversation: vbnet Copy Edit You: What's the weather in Delhi?

AI: calling tool getWeather

AI: Tool result: It's 32°C and sunny in Delhi.

You: Thanks! 🧪 Example Use Cases 🤖 Build AI agents that auto-call APIs/tools.

🔧 AI DevOps assistants that run shell commands.

🛠️ Product chatbot that uses backend tools to retrieve answers.

📊 AI dashboards with real-time tool access.

🔌 AI wrappers over microservices.

📄 File: index.js — Explained Section What it does dotenv/config Loads your .env variables (Gemini key) readline/promises Creates async CLI interface GoogleGenAI Initializes Gemini API client Client/SSEClientTransport Connects to MCP server listTools() Gets tool list + schemas generateContent() Sends prompt & tools to Gemini functionCall check If Gemini wants to call a tool mcpClient.callTool() Executes that tool via MCP chatHistory.push() Maintains back-and-forth memory chatLoop() Main loop that manages the entire process

🛠️ Troubleshooting Issue Fix 401 Unauthorized Check your Gemini API key Tool returns undefined Ensure the tool is properly defined in MCP 503 from Gemini Too many requests — retry later CLI freezes Use CTRL+C and restart SSE connection error Ensure MCP server is running at http://localhost:3001/sse

📈 Future Improvements 🧠 Multi-turn memory storage

💾 Save conversations locally

🌐 WebSocket support for real-time updates

🛡️ Authentication with OAuth for tools

📦 NPM package version of the agent

👨‍💻 Contributing Feel free to open issues or pull requests if you’d like to improve this project.

📝 License This project is open-source under the MIT License.

🙌 Acknowledgments Google Gemini

Model Context Protocol

OpenAI Function Calling inspiration

Community contributors