DI

This MCP server lets your AI assistants access decisions hosted in IBM Decision Intelligence or IBM Automation Decision Services, so you can plug governance, explainable outcomes, and automated decisions into conversational workflows and agents.

How to use

You run the MCP server locally or remotely, then have your MCP client discover and execute deployed decision services as MCP tools. The server supports both standard input/output (stdio) and HTTP transports, so you can work with local development environments or remote runtimes. Use an MCP client to start a session, browse available decision services, and invoke a specific decision for a given context. The server handles translating MCP tool invocations into calls to Decision Intelligence or Automation Decision Services and returns the resulting decisions and explanations.

To connect, start an MCP server instance that points to your decision runtime and authenticates with the appropriate credentials. Once running, your MCP client can enumerate the exposed decision services as tools and invoke them just like built-in operations.

How to install

Prerequisites you need on your machine before starting:

Install or run the MCP server using the commands shown for your target deployment.

Option A — IBM Decision Intelligence (Decision Intelligence runtime via API key) is exposed over a REST runtime URL. Start the MCP server with your DI API key and the Decision Intelligence runtime URL.

npx -y di-mcp-server --di-apikey <YOUR_DI_API_KEY> --url https://mytenant.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1

Additional configurations and details

The MCP server can be configured to connect to Decision Intelligence or Automation Decision Services. You can run it in stdio mode (default) or HTTP mode, and you can specify authentication methods such as API keys, Zen API keys, or basic credentials. The following examples demonstrate common usage patterns.

# Inline example: IBM Decision Intelligence with API key and runtime URL
npx -y di-mcp-server --di-apikey <DI_API_KEY> --url https://<TENANT_NAME>.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1

# Inline example: Automation Decision Services with Zen API key authentication
npx -y di-mcp-server --authentication-mode zenapikey --zen-username <ZEN_USERNAME> --zen-apikey <ZEN_API_KEY> --url https://<ADS_HOSTNAME>/ads/runtime/api/v1

# Inline example: Automation Decision Services with basic authentication
npx -y di-mcp-server --authentication-mode basic --basic-username <BASIC_USERNAME> --basic-password <BASIC_PASSWORD> --url https://<ADS_HOSTNAME>/ads/runtime/api/v1

Notes on transport and deployment

The MCP server supports both stdio and HTTP transports. Use stdio for local development and debugging, or HTTP when integrating with MCP clients that communicate over HTTP. For HTTP, the server acts as a bridge between the MCP client and the decision runtime.

Environment variables shown in examples provide credentials and host information. Replace placeholders with your actual values before running commands.

Tool naming and dynamic updates

Tools (decision services exposed as MCP operations) are discovered from the deployed decision services. You can customize tool names if needed by adjusting metadata on the decision runtime, ensuring names meet your host’s constraints.

The server can automatically detect tool changes and notify connected MCP clients so you stay in sync as decisions evolve.

Security considerations

Protect authentication credentials (API keys, Zen credentials, or basic user/password) and restrict runtime URLs to trusted networks. Use secure channels (HTTPS) for all remote communications.

Available tools

DecisionServices

Exposed Decision Intelligence decision services as MCP tools that you can invoke from your MCP client.