IBM Watsonx.data Intelligence MCP Server

Overview

IBM Watsonx.data Intelligence MCP Server is an open-source implementation of the Model Context Protocol (MCP) designed to act as a bridge between MCP-enabled clients (local or remote) and IBM Watsonx.data / Data Intelligence services. It provides a modular, secure and scalable server that forwards MCP interactions to IBM Data Intelligence runtimes, enabling tools like Claude Desktop, VS Code Copilot, Watsonx Orchestrate and IBM Bob to interact with enterprise data services via a standardized protocol.

The server supports multiple transport modes (stdio, HTTP, HTTPS), flexible authentication for SaaS and CPD deployments, and a pluggable set of “tools” (actions and helpers) used by models during prompt execution. The project repository and documentation are available at https://github.com/IBM/data-intelligence-mcp-server.

Features

• MCP protocol server compatible with MCP clients (stdio and HTTP transports)
• Multiple deployment modes: local stdio, HTTP, and HTTPS
• Integration with IBM Data Intelligence SaaS and CPD (Cloud Pak for Data)
• Environment-driven configuration (API key, service URL, mode)
• Pluggable “tools” and prompt templates (see TOOLS_PROMPTS.md)
• Logging and optional file-based log path
• Extensible for additional data-intelligence capabilities and custom tools

Installation / Configuration

Prerequisites:

• Python 3.11+
• For uvx mode: uvx/uv CLI (https://docs.astral.sh/uv/getting-started/installation/)
• Target Data Intelligence: SaaS or CPD 5.2.1+

Install from PyPI:

pip install ibm-watsonx-data-intelligence-mcp-server

Run as an HTTP server:

ibm-watsonx-data-intelligence-mcp-server --transport http --host 0.0.0.0 --port 3000

Run as HTTPS server:

• See detailed HTTPS configuration in the repository (readme_guides/SERVER_HTTPS.md).
• Provide TLS certificate and key via environment variables or configuration file (see table below).

Run with uvx (stdio transport, recommended for local desktop clients):

uvx ibm-watsonx-data-intelligence-mcp-server --transport stdio

Common environment variables

Adjust these variables before launching the server. For CPD installations you may also need to pass a username header when clients call the HTTP endpoint.

Tools and prompt templates

• Supported tools and example prompts are defined in TOOLS_PROMPTS.md in the repository.
• If an MCP client cannot register prompt templates programmatically, use PROMPTS_TEMPLATE_SAMPLES/ for manual templates.

Client configuration examples

Claude Desktop (stdio, recommended for local use)

{
  "mcpServers": {
    "wxdi-mcp-server": {
      "command": "uvx",
      "args": ["ibm-watsonx-data-intelligence-mcp-server", "--transport", "stdio"],
      "env": {
         "DI_SERVICE_URL": "https://api.dataplatform.cloud.ibm.com",
         "DI_APIKEY": "<data-intelligence-api-key>",
         "DI_ENV_MODE": "SaaS",
         "LOG_FILE_PATH": "/tmp/di-mcp-server-logs"
      }
    }
  }
}

Claude Desktop or other clients (remote HTTP/HTTPS)

• For SaaS:

{
  "mcpServers": {
    "wxdi-mcp-server": {
      "url": "https://mcp.example.com",
      "type": "http",
      "headers": { "x-api-key": "<your-cloud-saas-api-key>" }
    }
  }
}

• For CPD:

{
  "mcpServers": {
    "wxdi-mcp-server": {
      "url": "https://mcp.cpd.local",
      "type": "http",
      "headers": {
        "x-api-key": "<cpd-api-key>",
        "username": "<cpd-username>"
      }
    }
  }
}

VS Code Copilot, Watsonx Orchestrate and IBM Bob: configure servers similarly to the above JSON snippets, pointing to the MCP server URL or launching via stdio when supported.

Available Tools / Resources

• Repository: https://github.com/IBM/data-intelligence-mcp-server
• Prompt templates and samples: PROMPTS_TEMPLATE_SAMPLES/
• Tools and sample prompts: TOOLS_PROMPTS.md
• HTTPS server setup guide: readme_guides/SERVER_HTTPS.md
• Integration articles and walkthroughs:
  • Integrating Claude with Watsonx Data Intelligence
  • Watsonx Orchestrate + Data Intelligence
  • IBM Bob + Data Intelligence (Links available in the project README and documentation directories.)

Use Cases

1. Local developer testing with Claude Desktop

   • Run the MCP server in stdio mode via uvx to let Claude invoke it directly without an external web service. This simplifies local development and debugging while keeping Data Intelligence calls authenticated via DI_APIKEY.

2. Hosted MCP gateway for enterprise tools

   • Deploy the server as an HTTPS endpoint behind a load balancer. Configure enterprise MCP clients to call the HTTPS URL with x-api-key headers. The MCP server centralizes access to Watsonx.data and enforces consistent prompt/tooling behavior.

3. Automation pipelines with Watsonx Orchestrate

   • Use the MCP server to expose Data Intelligence tools as callable MCP “tools” for workflows. Orchestrate can then invoke these tools via MCP to run data queries, generate context-rich prompts, or orchestrate data jobs.

4. Hybrid CPD integration

   • For Cloud Pak for Data (CPD) environments, run the MCP server inside the same network and configure client headers (username and x-api-key) to authenticate and forward calls to the CPD runtime.

Notes and next steps

• Inspect TOOLS_PROMPTS.md to understand the available tool actions and example prompt templates.
• For production HTTPS deployments, follow the repository HTTPS guide to configure certificates and secure headers.
• Log files and environment-driven configuration make it straightforward to incorporate this server into CI/CD pipelines or containerized deployments (Docker/Kubernetes).