MCP Research Project

This project consists of two main components: a research server that handles paper retrieval and a chatbot client that interacts with the research server using the Model Context Protocol (MCP).

Project Structure

mcp_project/
├── research_server.py    # Server component for paper retrieval
├── mcp_chatbot.py       # Client chatbot component
├── .venv/               # Virtual environment directory
└── papers/              # Directory for downloaded papers

Server Component (research_server.py)

Prerequisites

• Python 3.x
• UV package manager
• Node.js and NPM (for the inspector)

Setup Instructions

1. Navigate to the project directory:

cd mcp_project_prompt_resources

2. Initialize with UV:

uv init

3. Create and activate virtual environment:

uv venv
source .venv/bin/activate

4. Install required dependencies:

uv add mcp arxiv
uv pip install -r requirements.txt

Running the Server

1. Launch the inspector:

npx @modelcontextprotocol/inspector uv run research_server.py

2. When prompted to install packages, type y

3. Configure the Inspector:

   • Open the provided inspector address in your browser
   • Under configuration, specify the Inspector Proxy Address
   • Follow the Inspector UI Instructions for detailed configuration steps

Accessing Papers

• Navigate to File > Open > L4 > mcp_project
• Papers will be stored in the papers/ directory

Stopping the Server

• Use Ctrl+C in the terminal to stop the inspector

Client Component (mcp_chatbot.py)

Prerequisites

• Python 3.x
• UV package manager
• Active virtual environment from server setup

Setup Instructions

1. Navigate to the project directory:

cd mcp_project_prompt_resources

2. Activate the virtual environment:

source .venv/bin/activate

3. Install additional dependencies:

uv add anthropic python-dotenv nest_asyncio

Running the Chatbot

1. Start the chatbot:

uv run mcp_chatbot.py

2. Interact with the chatbot through the command line
3. Type quit to exit the chatbot

Accessing Papers

• Navigate to File > Open > L5 > mcp_project to access downloaded papers

Environment Variables

The chatbot requires the following environment variables:

• ANTHROPIC_API_KEY: Your Anthropic API key for Claude access

Notes

• Ensure the server is running before starting the chatbot
• Keep track of API usage and paper downloads
• The server and client components should be run in separate terminals
• Make sure to properly close both applications when done

Resources and Prompt Templates

Paper Storage Structure

The research server organizes papers in the following structure:

• papers/ - Root directory for all research papers
  • {topic_name}/ - Folders for each research topic (e.g. ai_interpretability, llm_reasoning)
    • papers_info.json - JSON file containing metadata about papers in that topic

MCP Resources

Resources in MCP are read-only data endpoints similar to GET endpoints in REST APIs. The research server exposes two main resources:

1. List of available topic folders:

   • URI: papers://folders
   • Returns list of available research topics

2. Papers information under a specific topic:

   • URI: papers://{topic}
   • Returns paper metadata for the specified topic
   • Dynamic URI where {topic} is specified at runtime

Prompt Templates

The server provides customizable prompt templates using the @mcp.prompt() decorator. These templates can be used to generate structured prompts for specific use cases like paper searches.

Updated Chatbot Commands

The chatbot supports the following special commands:

Resource Commands

• @folders - List all available research topics
• @topic - Get papers information for a specific topic

Prompt Commands

• /prompts - List all available prompt templates
• /prompt <name> <arg1=value1> - Execute a specific prompt with arguments

Implementation Details

The chatbot has been updated with new methods to handle these features:

1. connect_to_server:

   • Requests available resources and prompt templates
   • Stores resource URIs and prompt names

2. chat_loop:

   • Processes special commands (@folders, @topic, /prompts, /prompt)
   • Routes commands to appropriate handlers

3. New Methods:

   • get_resource: Handles resource requests
   • execute_prompt: Processes prompt template execution
   • list_prompts: Shows available prompt templates

Remember to check the complete implementation in the mcp_project folder for detailed usage examples.

Make sure to interact with the chatbot. Here are some query examples:

• @folders
• @ai_interpretability
• /prompts
• /prompt generate_search_prompt topic=history num_papers=2