Building a Watsonx.ai Chatbot Server with MCP in Python

What is Watsonx Chatbot Server?

Watsonx Chatbot Server is a production‑ready MCP server that exposes IBM Watsonx.ai LLM inference as a standardized “chat” tool. It lets any MCP‑compatible client (e.g., Claude Desktop, custom Python clients) invoke Watsonx.ai responses through the Model Context Protocol.

How to use Watsonx Chatbot Server?

Set up Python 3.9+, a virtual environment, and install python-dotenv, ibm-watsonx-ai, and mcp[cli]. Store your Watsonx.ai API key, URL, and project ID in a .env file. Run mcp dev server.py for live‑reloading development with an Inspector UI at http://localhost:6274/, or run python server.py for direct STDIO mode. Use the sample client script to call the chat tool.

Key features of Watsonx Chatbot Server

• Exposes Watsonx.ai LLM inference as an MCP “chat” tool.
• Provides a dynamic greeting resource via @mcp.resource.
• Includes an optional symptom‑assessment prompt template.
• Supports live reloading and Inspector testing with mcp dev.
• Communicates over STDIO transport with any MCP client.
• Defaults to the ibm/granite-13b-instruct-v2 model.

Use cases of Watsonx Chatbot Server

• Building an internal helpdesk bot that responds to employee queries.
• Creating a public chatbot API that works with any MCP‑aware frontend.
• Rapidly prototyping medical symptom assessment assistants.
• Providing a reusable, modular LLM endpoint for multiple client applications.

FAQ from Watsonx Chatbot Server

What credentials are required to run the server?

You need an IBM Cloud Watsonx.ai API key, service URL, and project ID. Store them in a .env file as WATSONX_APIKEY, WATSONX_URL, and PROJECT_ID.

How do I install the dependencies?

Use pip install -r requirements.txt after creating a virtual environment. The required packages are python-dotenv>=0.21.0, ibm-watsonx-ai==1.3.8, and mcp[cli]>=1.6.0.

What transport does the server use?

The server uses STDIO transport for MCP communication. It runs as a command‑line process and exchanges JSON‑RPC messages over standard input/output.

How can I test the server without writing a client?

Run mcp dev server.py to start the development mode. Open http://localhost:6274/ in a browser to use the MCP Inspector for listing tools, calling the chat tool, and testing resources and prompts.

What model is used by default, and can I change it?

The default model is ibm/granite-13b-instruct-v2. You can override it by setting the MODEL_ID environment variable in your .env file.