mcp-chatbot

A modular, async research assistant that combines Anthropic Claude 3 with the Model Context Protocol (MCP), delivering on‐demand literature search and summarisation for academics and engineers.

---

1. Project Structure

mcp-chatbot/
├── Dockerfile
├── pyproject.toml
├── uv.lock
├── README.md
├── server_config.json
├── research_server.py
├── papers/                  # Cached paper metadata by topic
├── mcp_chatbot/
│   ├── __init__.py
│   ├── cli.py               # Typer-based CLI
│   └── core.py              # Main chatbot engine
└── tests/
    └── test_core.py

2. Quick Start

2.1. Clone the Repository

git clone https://github.com/mctrinh/mcp-chatbot.git
cd mcp-chatbot

2.2. Install Dependencies

Install uv (recommended)

# Git Bash or WSL on Windows, doesn't work in standard Command Prompt or PowerShell
curl -LsSf https://astral.sh/uv/install.sh | sh

# Scoop (Windows)
scoop install uv

# Chocolatey (Windows - Administrator Command Prompt - Recommended)
choco install uv
uv --version

Install Python packages in project.dependencies in pyproject.toml

pip install -e .

3. Build and Run with Docker

# Build image
docker build -t mcp-chatbot:0.1 .

# Run server and CLI (ports 8001 and 8000)
docker run --rm -it -p 8001:8001 -p 8000:8000 mcp-chatbot:0.1

4. Run Without Docker (Local Dev)

# Install dependencies
uv pip install -e .[dev]

# Start the research server (MCP tool)
python research_server.py

# In a new terminal, launch the chatbot CLI
mcp-chatbot run

5. Try the Chatbot

5.1. REPL Mode

python -m mcp_chatbot.cli run

Or using the installed script:

mcp-chatbot run

Once inside the REPL (Read-Eval-Print Loop), you can interact with the chatbot directly by typing commands or queries. Example commands:

/prompts            # list Claude prompts
@folders            # list downloaded paper topics
AI alignment        # ask anything – Claude decide whether to invoke tools

5.2. One-shot Query

mcp-chatbot once "What are the latest trends in diffusion models?"

6. Configuration (Optional)

Environment variables and server_config.json control model and ports:

export ANTHROPIC_MODEL="claude-3-opus-20240229"
export RESEARCH_PORT=8001
export PAPER_DIR=./papers

7. Testing

# Installs pytest, coverage, etc.
uv pip install -e .[dev]

# Run unit tests
pytest -q

# With coverage (optional)
pytest --cov=mcp_chatbot

8. Road map

• Research MCP server with search_papers and extract_info (done)

• Tool usage via Claude 3 (done)

• Prompt orchestration (done)

• Vector search over stored papers (Faiss / Chroma)

• Web UI using FastAPI + React

• GitHub Actions for CI/CD

9. License

MIT License. Copyright © 2025.

10. Current Issues

Issues occur when running mcp-chatbot run

• ⚠ Could not connect to server 'fetch': Method not found
• ⚠ Could not connect to server 'filesystem': Method not found