Paper Intelligence

A local MCP server for intelligent paper/PDF management. Convert PDFs to markdown, then search them with hybrid grep + semantic search. Designed for token efficiency: search first, read only what you need.

🚀 Quick Start

1. Install UV (one-time setup)

curl -LsSf https://astral.sh/uv/install.sh | sh

2. Add to Your MCP Client

Claude Code CLI:

claude mcp add paper-intelligence -- uvx paper-intelligence@latest

VS Code:

code --add-mcp '{"name":"paper-intelligence","command":"uvx","args":["paper-intelligence@latest"]}'

That's it! uvx handles everything automatically. Using @latest ensures you always get the newest version.

🔌 MCP Client Integration

{
  "mcpServers": {
    "paper-intelligence": {
      "command": "uvx",
      "args": ["paper-intelligence@latest"]
    }
  }
}

{
  "mcpServers": {
    "paper-intelligence": {
      "command": "uvx",
      "args": ["paper-intelligence@latest"]
    }
  }
}

{
  "mcpServers": {
    "paper-intelligence": {
      "command": "uvx",
      "args": ["paper-intelligence@latest"]
    }
  }
}

✨ Features

• PDF to Markdown — High-accuracy conversion using Marker
• Hybrid Search — Combined grep (exact/regex) + semantic RAG search
• Token Efficient — Search papers instead of reading entire documents
• GPU Acceleration — MPS (Apple Silicon) and CUDA support
• Self-Contained — Each paper gets its own directory with all data
• Header Context — Search results show document structure (e.g., "Methods > Data Collection")

📖 MCP Tools

search

Search one or more explicitly requested PDFs, processed paper directories, or library directories with grep, rag, or hybrid mode. Direct PDF searches never inspect sibling files or directories.

Parameters:

• query (string): Text, regex, or semantic query
• sources (array): PDF paths, paper directories, or an explicitly selected library directory
• mode (string, optional): "grep", "rag", or "hybrid" (default: hybrid)
• top_k (integer, optional): Number of results (default: 5)
• regex (boolean, optional): Treat the grep query as a regex (default: false)

A new or incomplete PDF is converted, indexed, and embedded in the background because this normally takes 1–3 minutes, longer than the roughly 30-second deadline used by many MCP clients. The first call returns promptly:

{
  "success": true,
  "status": "processing",
  "message": "First-use processing ... is continuing in the background.",
  "processing": [{
    "paper_dir": "/path/to/paper",
    "retry_after_seconds": 30,
    "next_step": "Call get_paper_info ..."
  }]
}

Processing continues after that response. Poll get_paper_info using paper_dir; when it reports status: "ready", retry the original search. Already-processed grep searches do not initialize the semantic model. RAG and hybrid searches initialize it when needed. Search no longer performs an unconditional remote-library sync, which previously allowed a local query to block for up to five minutes.

get_paper_info

Check a paper's processing state without loading the embedding model. Pass either the paper directory returned by search or the original PDF path.

Statuses are queued, processing, ready, incomplete, or failed. Failed responses include the background job's error message. A ready response includes artifact presence, metadata, and a lightweight local chunk count when available.

📊 Example Output

Search Result

{
  "source": "attention-is-all-you-need.md",
  "line_number": 142,
  "header_path": "Model Architecture > Attention",
  "content": "An attention function can be described as mapping a query and a set of key-value pairs to an output...",
  "score": 0.89
}

🎯 Typical Workflow

1. Process a paper:

   Process the PDF at ~/Downloads/transformer-paper.pdf

2. Search across papers:

   Search for "positional encoding" in my papers

3. Read specific sections:

   Show me the Methods section from the transformer paper

The agent reads search results (a few hundred tokens) instead of entire papers (tens of thousands of tokens).

🛠️ Installation Options

# Install with pip
pip install paper-intelligence

# Or run directly with uvx (no install needed)
uvx paper-intelligence@latest

pip install "paper-intelligence @ git+https://github.com/Strand-AI/paper-intelligence.git"

git clone https://github.com/Strand-AI/paper-intelligence.git
cd paper-intelligence

# Create virtual environment
python3.11 -m venv .venv
source .venv/bin/activate

# Install in development mode
pip install -e ".[dev]"

# Run the server
python -m paper_intelligence.server

{
  "mcpServers": {
    "paper-intelligence": {
      "command": "python",
      "args": ["-m", "paper_intelligence.server"],
      "cwd": "/path/to/paper-intelligence"
    }
  }
}

# Unit tests (fast)
pytest tests/test_markdown_parser.py

# Integration tests (slow, requires ML models)
pytest tests/test_integration.py -v

🔧 Debugging

Use the MCP Inspector to debug the server:

npx @modelcontextprotocol/inspector uvx paper-intelligence@latest

🆘 Troubleshooting

"env": {
  "PYTHONIOENCODING": "utf-8"
}

🏗️ Technical Stack

🙏 Acknowledgments

• Marker for excellent PDF conversion
• LlamaIndex for the RAG framework
• ChromaDB for the vector database
• FastMCP for the MCP server framework

📄 License

MIT — see LICENSE for details.