IBM Data Cataloging MCP server

Key capabilities

Before you begin

Before you enable and use the MCP server within the IBM Data Cataloging environment, ensure that your system meets the following requirements. These prerequisites are necessary to authenticate with your cluster, enable the MCP components, and verify the available service endpoints.

After verifying the prerequisites, log in to your OpenShift cluster by using CLI.

Setting up the MCP server

To enable MCP server capabilities, ensure that the Data Catalog has the required features that are enabled for the MCP server.

DCS_NS=ibm-data-cataloging

oc -n ${DCS_NS} patch SpectrumDiscover $(oc -n ${DCS_NS} get spectrumdiscover -o jsonpath='{.items[*].metadata.name}') \
  --type=merge -p '{
    "spec": {
      "enabled_features": {
        "natural-language-queries": {
          "enabled": true
        }
      }
    }
  }'

Once the patch is successfully applied, the SpectrumDiscover loads the MCP server with new capabilities.

Verifying that MCP server is running

After enabling the features, verify that the MCP server pods are up and running.

Run the following command:

oc -n ${DCS_NS} get pod -l 'app=isd,role=dcs-ai-mcp-server'

Example output:

isd-dcs-ai-mcp-server 1/1 Running 0 3h

If you see pods in the Running state, the MCP server is successfully up and operational.

Connecting to MCP server

Once the MCP server is deployed and prerequisites are met, you can retrieve the service endpoints to enable LLMs or interact with its tools. The MCP server exposes both SSE and HTTP endpoints for integration.

Retrieve MCP endpoints on Linux/macOS

oc get routes -n ${DCS_NS} -o json \
  | jq -r '.items[] 
  | select(.metadata.name | test("mcp")) 
  | "Route: \(.metadata.name)\nHost: \(.spec.host)\nSSE Endpoint: https://\(.spec.host)/mcp\nHTTP Endpoint: https://\(.spec.host)/mcp/http\n"'

Example output:

Route: dcs-mcp-route
Host: dcs-mcp-route-ibm-data-cataloging.apps.com
SSE Endpoint: https://dcs-mcp-route-ibm-data-cataloging.ibm.com/mcp
HTTP Endpoint: https://dcs-mcp-route-ibm-data-cataloging.com/mcp/http

The command outputs -

• Route: The OpenShift route for the MCP server.
• Host: The host URL assigned by OpenShift.
• SSE Endpoint: The URL to connect through server-sent events.
• HTTP Endpoint: The URL to connect through standard HTTP requests.

  This connection provides the necessary endpoints to integrate LLMs or other clients with the MCP server.

Initialization

Example of how to initialize the MCP server:

curl -i -X POST \
  <MCP http endpoint> \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "initialize",
    "params": {}
  }' --insecure

Replace all placeholder values before sending the request.

Example Output:

HTTP/1.1 200 OK
access-control-allow-headers: Content-Type
access-control-allow-methods: GET, POST, OPTIONS
access-control-allow-origin: *
cache-control: no-cache, no-transform
content-type: text/event-stream
mcp-session-id: xxxxxxxxxxxxxxxxxxxx
date: Tue, 16 Dec 2025 19:17:50 GMT
transfer-encoding: chunked
set-cookie: xxxxxxxxxxxxxxxxxxxxxxxx; path=/; HttpOnly; Secure; SameSite=None

Save the mcp-session-id, it is necessary to send requests.

List Tools

curl -X POST \
  <MCP http endpoint> \
  -H "Content-Type: application/json" \
  -H "Accept: application/json, text/event-stream" \
  -H "mcp-session-id: <mcp-session-id>" \
  -d '{
    "jsonrpc": "2.0",
    "id": 1,
    "method": "tools/list",
    "params": {}
  }' \
  --insecure

Replace all placeholder values before sending the request.

MCP server - Tools

The following section describes the new tools that are introduced in the MCP server, which is designed to improve interaction with data sources, manage credentials, policies, and tagging. Each tool serves a specific purpose and can be used independently or together to optimize data management and automation workflows.

dcs_set_credentials

The Set Credentials tool allows the user to configure the credentials that are required to connect the MCP server to the Data Cataloging Service.

Tool name	Arguments	Request
dcs_set_credentials	DCS URL, Username, Password	curl -X POST <mcp http endpoint> -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "mcp-session-id: <mcp-session-id>" -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "dcs_set_credentials", "arguments": { "cluster_url": "<DCS URL>", "username": "user", "password": "password" } } }' --insecure
	DCS URL, Authentication token	curl -X POST <mcp http endpoint> -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "mcp-session-id: <mcp-session-id>" -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "dcs_set_credentials", "arguments": { "cluster_url": "<DCS URL>", "token": "<token>" } } }' --insecure

Usage: It is used when starting a new session or when you need to update your connection credentials.

Replace all placeholder values before sending the request.

Obtaining a username and password:

You can retrieve the user and password by using the following commands:

macOS

#User
oc get secret keystone -n ${DCS_NS} -o jsonpath="{.data.user}" | base64 -D; echo

#Password
oc get secret keystone -n ${DCS_NS} -o jsonpath="{.data.password}" | base64 -D; echo

Linux

#User
oc get secret keystone -n ${DCS_NS} -o jsonpath="{.data.user}" | base64 --decode; echo

#Password
oc get secret keystone -n ${DCS_NS} -o jsonpath="{.data.password}" | base64 --decode; echo

Obtaining DCS URL:

Retrieve the DCS URL by using the following command:

DCS_CONSOLE=$(oc -n ${DCS_NS} get route console -o jsonpath='{"https://"}{.spec.host}')
echo $DCS_CONSOLE

Obtaining token:

Retrieve the token by using the following command:

curl -k -I -u <username>:<password> ${DCS_CONSOLE}/auth/v1/token | grep x-auth-token

Response:

x-auth-token: <token>

dcs_file_search

The File Search tool enables you to search across cataloged data sources by using natural language queries. It automatically converts user input into structured queries that retrieve relevant files, datasets, or metadata entries.

This capability allows you to find information quickly without needing to understand query syntax or underlying database languages. By bridging natural language with structured search, File Search improves accessibility for both technical and non-technical users, and accelerates data discovery.

Tool name	Argument	Request
dcs_file_search	Query	curl -X POST <mcp http endpoint> -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "mcp-session-id: <mcp-session-id>" -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "dcs_file_search", "arguments": { "query": "SELECT * FROM METAOCEAN LIMIT 1" } } }' --insecure

Replace all placeholder values before sending the request. Replace the query as needed.

Example prompt:

Show all CSV files containing customer transactions from 2023

The tool retrieves matching entries from the registered data sources.

dcs_get_registered_tags

The Get Registered Tags tool retrieves a complete list of all tags that are registered in the cataloging system. It helps users and AI agents review existing classifications before creating or suggesting new ones, ensuring that tag definitions remain consistent and non-duplicative.

This tool is useful for maintaining a clean and standardized tagging structure across multiple datasets and departments.

Tool name	Request
dcs_get_registered_tags	curl -X POST <mcp http endpoint> -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "mcp-session-id: <mcp-session-id>" -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "dcs_get_registered_tags", "arguments": { } } }' --insecure

Replace all placeholder values before sending the request.

Example use case:

Before adding a new Financial tag, you can check whether a similar tag exists in the catalog.

dcs_get_recommend_tags

The Recommend Tags tool uses LLM-powered analysis to suggest relevant tags for cataloged files and datasets. By examining metadata and contextual information, the tool recommends tags that help classify data more accurately and consistently.

This feature significantly reduces the time that is spent on manual tagging and enhances catalog quality by ensuring that all datasets are properly labeled.

Tool name	Argument	Request
dcs_get_recommend_tags	Query	curl -X POST <mcp http endpoint> -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "mcp-session-id: <mcp-session-id>" -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "dcs_get_recommend_tags", "arguments": { "query": "SELECT * FROM METAOCEAN LIMIT 10" } } }' --insecure

Replace all placeholder values before sending the request. Replace the query as needed.

Example prompt:

Recommend me some tags to classify my CSV files

The tool retrieves tags recommendation based on the metadata.

dcs_create_tag

The Create Tag tool enables you to define and create new tags.

Tool name	Arguments	Request
dcs_create_tag	tagName, type	curl -X POST <mcp http endpoint> -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "mcp-session-id: <mcp-session-id>" -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "dcs_create_tag", "arguments": { "tagName": "TEST", "type": "VARCHAR(256)" } } }' --insecure

Replace all placeholder values before sending the request.

Example prompt:

Create the tags that you recommend me.

The tool creates tags based on the recommendations.

dcs_get_policies

The Get Policies tool retrieves and lists all existing cataloging policies within the system. Policies define automated rules or governance conditions that are applied to data, metadata, or tags.

This tool provides visibility into the active policies that govern cataloged assets. It also helps users understand which rules are being enforced and how they affect data classification or compliance workflows.

Tool name	Request
dcs_get_policies	curl -X POST <mcp http endpoint> -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "mcp-session-id: <mcp-session-id>" -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "dcs_get_policies", "arguments": { } } }' --insecure

Replace all placeholder values before sending the request.

Example use case:

A data steward can use this tool to view all current auto-tagging and compliance policies before creating new ones.

dcs_create_policy

The Create AutoTag Policy tool enables you to define and deploy new auto-tagging policies that automatically assign tags to cataloged assets based on predefined rules and AI insights. It streamlines governance by automating repetitive classification tasks and ensuring that policies remain consistent across data sources.

This tool currently supports auto-tag policy types, allowing AI models to dynamically apply tags based on metadata patterns or user-defined contextual rules.

Tool name	Arguments	Request
dcs_create_policy	polFilter, policyName, schedule, tags	curl -X POST <mcp http endpoint> -H "Content-Type: application/json" -H "Accept: application/json, text/event-stream" -H "mcp-session-id: <mcp-session-id>" -d '{ "jsonrpc": "2.0", "id": 1, "method": "tools/call", "params": { "name": "dcs_create_policy", "arguments": { "polFilter": "filetype='pdf'", "policyName": "TEST", "schedule": "NOW", "tags": {'tagtest':'valtest'} } }' --insecure

Replace all placeholder values before sending the request.

Example prompt:

Explore the catalog for security-related flaws and recommend me 5 different tags with proposed values that can be used for identify and classify the data on the catalog, make sure they are not duplicate.