Scholar MCP

A lightweight MCP server for AI-assisted academic paper analysis.

Motivation

Automates early-stage literature review by connecting PDF parsing with Claude via MCP.

Supports:

PDF ingestion
Abstract extraction
Multi-paper comparison

Features

read_pdf — Extract full text from a PDF file
extract_abstract — Extract the abstract section
compare_papers — Extract and prepare multiple papers for comparison
ping — Test connectivity
update_analysis_policy (with dry-run) — Demonstrates safe state mutation with optional simulation mode

Architecture

Claude Desktop communicates with Scholar MCP via JSON-RPC over stdio (MCP protocol).
The server registers tools and handles execution through structured request handlers.

The design separates:

Read-only tools (PDF extraction & comparison)
State-changing tools (policy updates with optional dry-run)

All tool executions are wrapped in safe error handling to prevent server disconnections.

Installation

git clone https://github.com/mkjun2016/scholar-mcp.git
cd scholar-mcp
npm install

Usage

Run locally:

node server.js

Configure in Claude Desktop:

{
  "mcpServers": {
    "scholar-mcp": {
      "command": "/absolute/path/to/node",
      "args": ["/absolute/path/to/scholar-mcp/server.js"]
    }
  }
}

Example Workflow

Call compare_papers with two PDF paths
Ask Claude to:
- Compare research problems
- Summarize methodological differences
- Identify key contributions
- Highlight research gaps

This enables rapid AI-assisted literature comparison.

Technical Notes

Built on JSON-RPC over stdio (MCP) with robust error handling to ensure reliable tool execution.

Demo

Example interaction using compare_papers with Claude Desktop:

Demo Screenshot

License

ISC

MCP Servers