MCP server for IBM Decision Intelligence and IBM Decision Intelligence Client Managed Software

flowchart LR
   client("AI Assistant <br><br>(Claude, Watsonx Orchestrate, etc.)") -- MCP --> server("Decision Intelligence<br>MCP Server")
   server -- HTTPS --> runtime("IBM Decision Intelligence <br>or IBM DI CMS <br><br>Decision Runtime")

This Model Context Protocol (MCP) server empowers AI assistants by accessing decisions from IBM Decision Intelligence or IBM Decision Intelligence Client Managed Software.

The MCP server is available as an npm package in the free npm registry at https://www.npmjs.com/package/di-mcp-server.

It supports both STDIO and streamable HTTP transports for local or remote deployments for supporting any MCP clients.

Resources

• Supercharge your AI Assistants with the decisioning capabilities of IBM Decision Intelligence: Step-by-step instructions on how to use the MCP server for IBM Decision Intelligence.
• Trusted automated decisions in Agentic application with IBM Decision Intelligence and MCP: This video shows how an agentic application can leverage automated decision services that are defined and hosted in the IBM Decision Intelligence platform to make accurate and explainable decisions.
• Enrich your watsonx Orchestrate chat experience with IBM Decision Intelligence: Step-by-step instructions on how to use the decisioning capability in IBM watsonx Orchestrate.

Getting started with the MCP server

You can use the MCP server available in the npm registry. To develop your own MCP server or contribute to the development, see Developing the MCP server.

You can run the MCP server with npx to expose each operation of the last deployed version of a decision service as an MCP tool.

For IBM Decision Intelligence

Create the MCP server by using decisions that are deployed in Decision Intelligence:

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

For IBM Decision Intelligence Client Managed Software (IBM DI CMS)

Zen API key authentication

Create the MCP server by using decisions that are deployed in IBM DI CMS that uses the Zen API key authentication:

npx -y di-mcp-server --authentication-mode zenapikey --zen-username <YOUR_ZEN_USERNAME> --zen-apikey <YOUR_ZEN_API_KEY> --url https://myads-hostname/ads/runtime/api/v1

Basic authentication

Create the MCP server by using decisions that are deployed in IBM DI CMS that uses the basic authentication:

npx -y di-mcp-server --authentication-mode basic --basic-username <YOUR_USERNAME> --basic-password <YOUR_PASSWORD> --url https://myads-hostname/ads/runtime/api/v1

Command line syntax

Syntax of the command line:

npx -y di-mcp-server [--authentication-mode <AUTHENTICATION_MODE>] <CREDENTIALS> --url <RUNTIME_BASE_URL> [--transport <TRANSPORT>] [--deployment-spaces <DEPLOYMENT_SPACES>] [--decision-service-ids <DECISION_SERVICE_IDS>] [--decision-service-poll-interval <DECISION_SERVICE_POLL_INTERVAL>]

where

• AUTHENTICATION_MODE (optional) is the authentication mode to access the decision runtime; either diapikey (default), zenapikey, or basic respectively for authenticating with the Decision Intelligence API key, Zen API key, or basic credentials (i.e. username and password)
• CREDENTIALS is one of the following options, depending on the chosen authentication mode:
  • For Decision Intelligence API key authentication: --di-apikey <DI_API_KEY> where DI_API_KEY is the API key to access the decision runtime for Decision Intelligence.
  • For Zen API key authentication: --zen-username <ZEN_USERNAME> --apikey <ZEN_API_KEY> where ZEN_USERNAME and ZEN_API_KEY are the Zen API key credentials to access the decision runtime for IBM DI CMS (see Authorizing HTTP requests by using the Zen API key)
  • For basic authentication: --basic-username <BASIC_USERNAME> --basic-password <BASIC_PASSWORD> where BASIC_USERNAME and BASIC_PASSWORD are the basic authentication credentials to connect to the decision runtime for IBM DI CMS.
• RUNTIME_BASE_URL is the base URL of the decision runtime REST API. For Decision Intelligence, its pattern is: https://<TENANT_NAME>.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1 where TENANT_NAME is the name of the tenant.
• TRANSPORT (optional) is the transport protocol, either stdio (default) or http.
• DEPLOYMENT_SPACES (optional) is a comma-separated list of deployment spaces to scan (defaults to development).
• DECISION_SERVICE_IDS (optional) If defined, comma-separated list of decision service ids to be exposed as tools
• DECISION_SERVICE_POLL_INTERVAL (optional) is the interval in seconds for polling decision services (default: 30, minimum: 1)

The following environment variables can be used in addition to the command line options.

Integrating decision services into AI applications

The MCP server for Decision Intelligence extends its capability by enabling AI applications, such as IBM watsonx Orchestrate and Claude, to discover and execute deployed decision services.

Integrating decision services into IBM watsonx Orchestrate

The article Enrich your watsonx Orchestrate chat experience with IBM Decision Intelligence demontrates how to integrate the capability to discover and execute decisions in an IBM watsonx Orchestrate agent by using the Decision Intelligence MCP server.

See the IBM watsonx Orchestrate integration page for passing sensitive configuration settings to the IBM Decision Intelligence MCP Server.

Integrating decision services into Claude Desktop

You can integrate decision services into Claude Desktop by adding the MCP server.

1. Locate the Claude Desktop configuration file.

   Find your Claude configuration directory:

   • macOS: ~/Library/Application\ Support/Claude/claude_desktop_config.json
   • Windows: %APPDATA%\Claude\claude_desktop_config.json
   • Linux: ${HOME}/.config/Claude/claude_desktop_config.json

2. Add the MCP server configuration to the configuration file.

   • In the configuration directory, edit or create claude_desktop_config.json:

     hljs language-json

     Copy

     {
         [..]
         "mcpServers": {
             "di-mcp-server": {
                 "command": "npx",
                 "args": [
                     "-y",
                     "di-mcp-server",
                     "--di-apikey",
                     "<APIKEY>",
                     "--url",
                     "https://<TENANT_NAME>.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1"
                 ]
             }
         }
         [..]
     }
     

   • Alternatively, you can use the APIKEY and URL environment variables to respectively specify the API key and the base URL of the decision runtime REST API:

     hljs language-json

     Copy

     {
         [..]
         "mcpServers": {
             "di-mcp-server": {
                 "command": "npx",
                 "args": ["-y", "di-mcp-server"],
                 "env": {
                     "DI_APIKEY": "<APIKEY>",
                     "URL": "https://<TENANT_NAME>.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1"
                 }
             }
         }
         [..]
     }
     

For more information, see https://modelcontextprotocol.io/quickstart/user.

Integrating decision services into Cursor

You can integrate decision services into Cursor by adding the MCP server.

1. In Cursor, click the cog wheel icon to open the Cursor settings.

2. Click Tools & Integration in the settings categories that are listed on the left.

   

3. Click + New MCP Server, to open Cursor's mcp.json configuration file.

   

4. Add a new MCP server entry.

   As for Claude Desktop, you can specify the API key and base URL of the decision runtime REST API using with one of the following methods:

   • Using command line arguments:
     hljs language-json

     Copy

     {
         [..]
         "mcpServers": {
             "di-mcp-server": {
                 "command": "npx",
                 "args": [
                     "-y",
                     "di-mcp-server",
                     "--di-apikey",
                     "<APIKEY>",
                     "--url",
                     "https://<TENANT_NAME>.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1"
                 ]
             }
         }
         [..]
     }
     

   • Using environment variables:
     hljs language-json

     Copy

     {
         [..]
         "mcpServers": {
             "di-mcp-server": {
                 "command": "npx",
                 "args": ["-y", "di-mcp-server"],
                 "env": {
                     "DI_APIKEY": "<APIKEY>",
                     "URL": "https://<TENANT_NAME>.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1"
                 }
             }
         }
         [..]
     }
     

For more information, see Installing MCP servers in the Cursor documentation.

Using custom tool names

When you integrate with MCP hosts, you might need to customize the tool names to meet specific requirements or limitations, such as:

• Maximum length restrictions
• Forbidden characters
• Naming conventions

Default tool naming algorithm

By default, tool names are generated in the following way:

1. Combines the decision service name with the operation ID: decisionServiceName operationID
2. Replaces spaces and forward slashes with underscores: decisionServiceName_operationID
3. Handles name collisions by using the decision service ID: use decisionServiceID_operationID if decisionServiceName_operationID already exists

Customizing tool names with the decision runtime REST API

If the default naming strategy doesn't meet the requirements of your MCP hosts, you can specify custom tool names by setting the mcpToolName.OPERATION_ID decision metadata:

{
  "map": {
    [..]
    "mcpToolName.OPERATION_ID": {
      "name": "mcpToolName.OPERATION_ID",
      "kind": "PLAIN",
      "readOnly": false,
      "value": "YourCustomToolName"
    }
    [..]
  }
}

where

• OPERATION_ID is the operation unique identifier
• YourCustomToolName is the desired tool name for the operation

Filtering exposed decision services

By default, the MCP server exposes all decision services from the specified deployment spaces as tools. You can filter which decision services are exposed using one of two methods:

Method 1: Filter by decision service IDs

Use the --decision-service-ids option to expose only specific decision services by their unique identifiers.

The decision service ID is defined in the decisionServiceId metadata of the deployed decision service.

Example configuration:

{
    "mcpServers": {
        "di-mcp-server": {
            "command": "npx",
            "args": [
                "-y",
                "di-mcp-server",
                "--di-apikey",
                "<APIKEY>",
                "--url",
                "https://<TENANT_NAME>.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1",
                "--decision-service-ids",
                "loan-approval,credit-scoring"
            ]
        }
    }
}

Method 2: Filter by MCP groups

Use the --mcp-groups option to expose decision services that belong to specific logical groups. This is useful for organizing decision services by domain, team, or use case.

The MCP groups are defined in the mcpGroups metadata of the deployed decision service as a comma-separated list.

Example configuration:

{
    "mcpServers": {
        "di-mcp-server": {
            "command": "npx",
            "args": [
                "-y",
                "di-mcp-server",
                "--di-apikey",
                "<APIKEY>",
                "--url",
                "https://<TENANT_NAME>.decision-prod-us-south.decision.saas.ibm.com/ads/runtime/api/v1",
                "--mcp-groups",
                "finance,risk-management"
            ]
        }
    }
}

Dynamic Tool Updates

For information about how the MCP server automatically detects and notifies clients of tool changes, see Dynamic Tool Updates.

Technical details

flowchart LR
    github["di-mcp-server github repository"] -- publish --> registry
    registry["NPM registry"] -- npx -y di-mcp-server--> server

    subgraph MCP Host 
        client["MCP Client"] <-- MCP/STDIO --> server("DI MCP Server")
    end

    server -- HTTPS --> runtime("Decision Runtime")

    subgraph id["Decision Intelligence<br>or Decision Intelligence Client Managed Software"]
        runtime
    end


    client <-- MCP/HTTP --> server2("DI MCP Server") -- HTTPS --> runtime

License

Apache 2.0

Notice

Copyright contributors to the IBM Decision Intelligence / Decision Intelligence Client Managed Software MCP Server project

Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License.You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.