LangGraph SQL Assistant

CareDesk is an AI-powered hospital receptionist application that integrates with a MySQL database via an MCP Database Server. In this example, it is configured for a hospital use case, but you can adapt it for your own domain such as retail, education, finance, or any other database-driven chatbot — by changing the system prompt, database schema, and tools configuration.

• Main entrypoint: agent.py
• Dependencies: requirements.txt
• System prompt: files in prompts/ (e.g., hospital_receptionist.txt) you can edit it for you
• MCP DB server: Node.js project providing tools via stdio transport

Requirements

• Python 3.10+
• Node.js 18+
• MySQL 8+ (or a compatible MySQL server)

Python dependencies

Install from requirements.txt:

pip install -r requirements.txt

Current pinned packages (see requirements.txt):

• python-dotenv
• pydantic
• langchain, langchain-openai, langchain-core
• langgraph
• langchain-mcp-adapters
• python-multipart
• rich

Environment configuration

Create a .env file in the project root. The agent expects:

# GitHub Models (used in agent.py)
GITHUB_TOKEN=ghp_your_github_token_here 

# Optional: override DB defaults
DB_HOST=localhost
DB_NAME=hospital
DB_PORT=3306
DB_USER=root
DB_PASSWORD=your_mysql_password

Notes:

• agent.py maps GITHUB_TOKEN to OPENAI_API_KEY and sets OPENAI_API_BASE=https://models.github.ai/inference to use GitHub Models.
• If you prefer a different OpenAI-compatible provider, adjust those envs in agent.py accordingly.

Prompts

Place role/system prompts under prompts/. In this example, it is configured for a hospital use case using the hospital_receptionist.txt system prompt, but you can replace it with your own role prompt for any domain such as retail, education, finance, or other database-driven chatbots.

• Default used: prompts/hospital_receptionist.txt
• If missing, agent.py falls back to a built-in system prompt.

MCP Database Server (MySQL)

agent.py launches a local MCP server over stdio. The code expects the built server at:

<mrepo>/mcp-database-server/dist/src/index.js

If you don’t have it yet, download and build it as follows (then place/rename the folder to mcp-database-server in the project root):

1. Clone the repository:

   git clone https://github.com/executeautomation/mcp-database-server.git
   cd database-server

2. Install dependencies:

   npm install

3. Build the project:

   npm run build

After building, ensure the folder name under this project root is exactly mcp-database-server so the path in agent.py resolves:

customerSupport/
├─ agent.py
├─ mcp-database-server/
│  └─ dist/src/index.js
└─ prompts/

If your cloned folder is named differently (e.g., database-server), either rename it to mcp-database-server or update the path in agent.py where mcp_server_path is set.

MySQL preparation

Ensure your MySQL instance is reachable and the credentials in .env are valid. The default database name is hospital. Seed your schema/data as appropriate for your use case.

Run the agent

From the project root:

python agent.py

You’ll see a welcome panel and a list of available MCP database tools on successful connection. Try queries like:

• "List tables"
• "Describe table patients"
• "Show doctors"

Type exit to quit.

Troubleshooting

• "Failed to get MCP tools":
  • Verify Node.js is installed and the server path exists at mcp-database-server/dist/src/index.js.
  • Make sure you ran npm install && npm run build inside the MCP server folder.
• Authentication errors with model:
  • Ensure .env contains a valid GITHUB_TOKEN and your network allows access to https://models.github.ai/inference.
• DB connection errors:
  • Confirm DB_HOST, DB_NAME, DB_PORT, DB_USER, DB_PASSWORD are correct and the database is reachable.

File references

• Agent: agent.py
• Requirements: requirements.txt
• Prompts directory: prompts/ (e.g., prompts/hospital_receptionist.txt)