Paper Research Helper

A LangChain + LangGraph research assistant that ingests academic papers and answers questions about them, exposed as an MCP server compatible with Cursor and Claude Desktop.

Features

• Paper ingestion — fetch PDFs from arXiv by ID, extract text, chunk and embed into a local FAISS vector store.
• Semantic search — search arXiv and Semantic Scholar for papers by keyword.
• QA over papers — ask natural-language questions answered with retrieved passages from ingested papers.
• MCP server — expose all capabilities as tools consumable by any MCP-compatible client.

Project Structure

paper_research_helper/
├── main.py                    # CLI entry point (serve / ingest / ask)
├── requirements.txt
├── .env.example
├── docs/
│   ├── architecture.md        # System design & data flow
│   └── getting_started.md    # Step-by-step setup guide
└── src/
    ├── adapters/              # External service connectors
    │   ├── arxiv.py           #   arXiv API
    │   ├── semantic_scholar.py #  Semantic Scholar API
    │   └── pdf.py             #   PDF text extraction
    ├── tools/                 # LangChain tools used by agents
    │   ├── search.py          #   Paper search tool
    │   ├── retrieval.py       #   Vector store retrieval tool
    │   └── summarize.py       #   LLM summarization tool
    ├── agents/                # LangGraph agent nodes
    │   ├── research_agent.py  #   Discovers & summarises papers
    │   └── qa_agent.py        #   RAG question-answering agent
    ├── graphs/                # LangGraph state graphs
    │   └── research_graph.py  #   Full research pipeline graph
    ├── pipeline/              # Ingestion orchestration
    │   └── ingestion.py       #   Fetch → chunk → embed → index
    └── mcp/                   # MCP server
        └── server.py          #   FastMCP server with 3 tools

Quick Start

1. Install dependencies

uv sync           # creates .venv and installs all dependencies

2. Configure environment

cp .env.example .env
# edit .env and set OPENAI_API_KEY at minimum

3. Ingest a paper

python main.py ingest --arxiv-id 2301.07041   # Attention Is All You Need (example)

4. Ask a question

python main.py ask "What problem does the transformer architecture solve?"

5. Start the MCP server

python main.py serve

Then add the server to your Cursor or Claude Desktop MCP config:

{
  "mcpServers": {
    "paper-research-helper": {
      "command": "python",
      "args": ["main.py", "serve"],
      "cwd": "/path/to/paper_research_helper"
    }
  }
}

MCP Tools

License

MIT