mochow13/google-scholar-mcp

A Model Context Protocol (MCP) server that provides Google Scholar search capabilities through a streamable HTTP transport. This project demonstrates how to build an MCP server with custom tools and integrate it with AI models like Google's Gemini.
Overview
This project consists of two main components:
- **MCP Server**: Provides Google Scholar search tools via HTTP endpoints
- **MCP Client**: Integrates with Google Gemini AI to process queries and call tools
Architecture
MCP Server Implementation
The server is built using the @modelcontextprotocol/sdk and implements:
- **Transport**: StreamableHTTPServerTransport for HTTP-based communication
- **Session Management**: Supports multiple simultaneous connections with session IDs
- **Tool System**: Extensible tool registration and execution framework
- **Error Handling**: Comprehensive error responses and logging
Available Tools
The server currently provides one main tool:
search_google_scholar
- **Description**: Search Google Scholar for academic papers and research
- **Parameters**: Configurable search parameters (query, filters, etc.)
- **Returns**: Structured search results with paper details
Transport Protocol
The server uses **StreamableHTTPServerTransport** which supports:
- **HTTP POST**: For sending requests and receiving responses
- **HTTP GET**: For establishing Server-Sent Events (SSE) streams
- **Session Management**: Persistent connections with unique session IDs
- **Real-time Notifications**: Streaming updates via SSE
Smithery
The server is now available in Smithery: Google Scholar Search Server
Installation
Installing via Smithery
To install google-scholar-mcp for Claude Desktop automatically via Smithery:
1. Clone the repository:
2. Install and build:
Running the Server
1. Start the MCP server:
The server will start on port 3000 and provide the following endpoints:
- POST /mcp - Main MCP communication endpoint
- GET /mcp - SSE stream endpoint for real-time updates
Server Features
- **Multi-session Support**: Handle multiple clients simultaneously
- **Graceful Shutdown**: Proper cleanup on SIGINT
- **Logging**: Comprehensive request/response logging
- **Error Handling**: Structured JSON-RPC error responses
Running the Client
The client demonstrates how to integrate the MCP server with Google's Gemini AI model.
1. Ensure you have a valid GEMINI_API_KEY and provide it with
2. Start the client:
3. The client will connect to the server and start an interactive chat loop
Client Features
Conversation Management
- **Persistent Context**: Maintains full conversation history across queries
- **Multi-turn Conversations**: Supports back-and-forth dialogue with context
- **Function Call Integration**: Seamlessly integrates tool calls into conversation flow
AI Integration
- **Gemini 2.5 Flash**: Uses Google's latest language model
- **Tool Discovery**: Automatically discovers and registers available MCP tools
- **Function Calling**: Converts MCP tools to Gemini function declarations