AI Research Assistant MCP

An MCP-powered AI research assistant for discovering, understanding, comparing, and organizing academic papers.

The server connects to arXiv for paper discovery and uses Gemini to perform research-oriented analysis. It is designed to help researchers move from finding papers to understanding their methods, limitations, and potential research gaps.

Features

🔎 Paper Discovery

Search arXiv using a research topic or keyword.

The server returns:

Paper title
Authors
Abstract
Publication date
Updated date
arXiv URL
PDF URL

📄 Paper Retrieval

Retrieve detailed information about a specific paper using its arXiv ID or URL.

📝 Paper Summarization

Generate a structured research-oriented summary covering:

Research problem
Proposed approach
Key results
Main contribution
Why the work matters

⚖️ Paper Comparison

Compare multiple papers across:

Research problem
Core method
Model/retrieval architecture
Dataset or evaluation
Results
Strengths
Weaknesses
Differences
Common findings
Research opportunities

🔬 Methodology Analysis

Extract and explain the methodology of a paper, including:

Research objective
Architecture
Main technique
Data used
Training/retrieval process
Evaluation method
Technical components

⚠️ Limitation Analysis

Analyze papers for:

Explicit limitations
Methodological weaknesses
Evaluation limitations
Dataset limitations
Generalization concerns
Computational concerns
Future research questions

The assistant distinguishes between limitations explicitly supported by the paper and potential research questions inferred from the available information.

💡 Research Gap Discovery

Given a research topic and multiple papers, identify potential research gaps and provide:

Gap
Evidence from the literature
Why the gap matters
Possible research question
Possible experiment
Expected contribution

The system is instructed not to present an unverified gap as an established fact.

💾 Research Library

Papers can be saved locally to create a personal research library.

Available operations include:

Save a paper
List saved papers
Search saved papers

Saved papers are stored in saved_papers.json.

Architecture

             AI Client
                │
                │ MCP
                ▼
      ┌─────────────────────┐
      │  AI Research        │
      │  Assistant Server   │
      └──────────┬──────────┘
                 │
         ┌───────┴────────┐
         │                │
         ▼                ▼
       arXiv            Gemini
    Paper Search      AI Analysis
         │                │
         └───────┬────────┘
                 ▼
          Research Insights

The server is implemented with FastMCP and exposes research capabilities as MCP tools.

MCP Tools

The current server provides:

search_papers() get_paper() summarize_paper() compare_papers() extract_methodology() extract_limitations() find_research_gaps() save_paper() list_saved_papers() search_saved_papers()

Technology Stack

Python 3.11+
FastMCP
Model Context Protocol (MCP)
arXiv API
Google Gemini API
Requests
JSON-based local storage

Project Structure

research-assistant/
│
├── src/
│   └── research_assistant/
│       └── __init__.py
│
├── server.py
├── pyproject.toml
├── README.md
├── .gitignore
└── .python-version

Setup

Install the project dependencies with:

uv sync

Set your Gemini API key as an environment variable:

GEMINI_API_KEY=your_api_key

Do not commit API keys or .env files to GitHub.

Running the Server

Run the MCP server locally:

uv run server.py

Testing with MCP Inspector

Run:

uv run fastmcp dev inspector server.py

The MCP Inspector can then be used to test the available tools and verify their responses.

Example Workflow

A typical research workflow can look like:

Research Topic ↓ Search Papers ↓ Select Relevant Papers ↓ Retrieve Paper Details ↓ Summarize Papers ↓ Compare Papers ↓ Analyze Methodology ↓ Analyze Limitations ↓ Identify Potential Research Gaps ↓ Save Important Papers

For example: text "Find papers about retrieval augmented generation and identify potential research gaps."

The assistant can search the literature, retrieve relevant papers, analyze them, and use the available paper information to identify potential research directions.

Research Focus

The project is being developed with a particular interest in Generative AI and Retrieval-Augmented Generation (RAG) research.

Potential future research workflows include studying:

CRAG
Self-RAG
RAG evaluation
Retrieval quality
Hallucination reduction
Semantic retrieval
Conflicting evidence
Research-paper comparison
Literature-gap discovery

Current Status

🚧 Active development

Current capabilities focus on:

Paper discovery → Paper analysis → Paper comparison → Research-gap exploration → Personal paper library

Future versions can extend this into deeper paper-level analysis, full-text research, retrieval over saved papers, and more advanced research workflows.

Author

Khushi Sonwane

Built as a hands-on project exploring Generative AI, MCP, academic research workflows, and RAG systems.

ai-research-assistant-mcp

PLUG IN

README