created by

Tamás András Péter 2026-08

ParaBank MCP Client & Server (Gemini API Integration)

A TypeScript implementation of the Model Context Protocol (MCP) integrated with Google's latest Gemini API (@google/genai). This project acts as an intelligent AI agent bridge: it dynamically discovers tools provided by a local ParaBank MCP server over standard input/output (stdio), converts the tool definitions into Gemini-compatible function declarations, and executes requested banking operations on demand.

🏗️ Architecture Overview

┌─────────────────────────────────────────────────────────────────────────────┐ │ MCP CLIENT │ │ (src/client/gemini.ts) │ │ │ │ 1. Starts MCP Server via Child Process (stdio) │ │ 2. Queries Server Capabilities via listTools() │ └──────────────┬──────────────────────────────────────────────▲───────────────┘ │ Convert MCP Tools │ Tool Output │ to Function Declarations │ JSON Payload ▼ │ ┌──────────────────────────────┐ ┌───────────────┴───────────────┐ │ GOOGLE GEMINI API │ │ MCP SERVER │ │ (gemini-3.6-flash) │ │ (dist/index.js) │ │ │ │ │ │ Evaluates User Prompt & │ │ Executes requested banking │ │ emits structured ├─────────────►│ action against ParaBank API │ │ FunctionCall object │ FunctionCall │ or local mocks │ └──────────────────────────────┘ Details └───────────────────────────────┘

Protocol Flow

1. Tool Discovery: The client spawns the compiled MCP server sub-process (dist/index.js) and fetches available tool definitions.

2. Declaration Mapping: Tool schemas from the MCP server are dynamically mapped to Google Gemini's standard FunctionDeclaration schema format.

3. Intent Recognition: User prompts are sent to gemini-2.5-flash. If Gemini determines a banking operation is required, it returns a functionCall request.

4. Execution over Stdio: The client intercepts the functionCall, invokes mcpClient.callTool() over the bidirectional stdio transport layer, and outputs the result.

Related MCP server: MifosX Self Service MCP

🛠️ Tech Stack

• Language: Node.js v24+ & TypeScript (ESM)

• AI Model Engine: Google Gen AI SDK (@google/genai)

• Protocol Standard: Model Context Protocol SDK (@modelcontextprotocol/sdk)

• Target Services: ParaBank Banking Operations

🚀 Getting Started

Prerequisites

• Node.js (v20+ or v24 recommended)

• A Gemini API key from Google AI Studio

1. Installation

Clone the repository and install the project dependencies:

git clone [https://github.com/your-username/parabank-MCP-Gemini.git](https://github.com/your-username/parabank-MCP-Gemini.git)
cd parabank-MCP-Gemini
npm install