Following our public preview earlier this year, we are excited to announce the 1.0.0 release IBM watsonx.data integration MCP Server. This MCP server connects your data integration workflows directly to the AI Agent of your choice, whether it's IBM Bob, Codex, Claude, or another compatible agent, so you can build, manage, and optimize your pipelines entirely through natural language.

The release is more than a stability milestone. It introduces a new planning workflow, expands discoverability to batch flows, adds a fourth out-of-the-box skill, strengthens on-premises support, and includes a round of usability improvements driven by what we learned during preview.

What's new in release

Agents now think before they act. The biggest behavioral change is that agents no longer jump straight to building. They draft an intent document first, sources, stages, transformations, targets, and wait for your review. Misunderstandings surface as sentences you can correct, not failed jobs you have to debug. First-attempt success goes up because the agent builds the pipeline you actually wanted.

Agents now see the full platform. Stage and property discovery, previously limited to streaming flows, now extends to batch. Agents enumerate available stages on demand and inspect each stage's required properties and preconditions, choosing configurations grounded in what the platform actually exposes rather than guessing from memory and producing malformed fields.

Agents now run reliably in real environments. Two preview pain points are resolved. The sandbox script execution tool times out far less often, so longer-running automations actually complete. And agents now reach every data center location without the connection issues that blocked some preview users.

Picture this

Here's what it looks like from your side of the conversation.

"Build a multi-source union flow with filtering and schema mapping, then run it once we're done."

Your agent reaches for the planning tool first. It drafts an intent document laying out the sources, the union and filter stages, the schema mapping, and the target. You see the whole plan before it builds anything, correct a mapping, approve it, and the agent constructs the multi-stage flow and runs it. What lands on your desk is a working pipeline and the script that built it.

"Inspect all the flows in my project and summarize my data flow architecture."

The agent reads across your flows and connections to assemble the full picture, then explains how your data moves end to end, grounded in your actual assets rather than a guess.

"Pull the logs for the failed job, tell me what broke, and fix it."

The agent fetches the failed run's logs, reasons through the root cause, and proposes a fix. Once you approve, it applies the change. The diagnosis and the fix both show up as code you can review before anything touches production.

"Swap an on premise data source with a new cloud data source across every flow in this project. Show me the plan first, then run it."

The agent lays out exactly which flows it will touch and how, waits for your go-ahead, then loops through each flow applying the change. Cautious about a bulk migration? An engineer can validate one swap manually, have the agent capture it as a reusable script, then run that script across every remaining flow. Either way, what would have been a week of careful edits becomes a reviewed plan and a single confirmation

Tools and skills at a glance

TheGA release ships with the following set of tools and skills:

Tools

• Intent document — the planning tool that drafts a reviewable pipeline plan before building.

• Stage and property discovery — for both streaming and batch flows.

• SDK documentation list and lookup — find and reference the right SDK methods on demand.

• Sandbox script execution — run SDK automation scripts in an isolated sandbox.

Skills

• Batch flows

• Streaming flows

• Platform

• Jobs

Getting started with the IBM watsonx.data integration MCP Server

Requirements

Python 3.12

Setup

There are three ways to run this MCP server. Choose the option that best fits your environment.

Option 1 — Using uvx (Recommended)

The fastest way to get started, with no installation required.

If you do not have uvx, install it first:

pip install uv

Verify it works:

uvx --python 3.12 --from ibm_watsonx_data_integration_mcp data-intg-mcp --help

MCP server configuration:

{
  "mcpServers": {
    "data-intg-mcp": {
      "command": "uvx",
      "args": [
        "--python",
        "3.12",
        "--from",
        "ibm_watsonx_data_integration_mcp",
        "data-intg-mcp"
      ],
      "env": {
        "WATSONX_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Option 2 — Using uv as a Tool

If you do not have uv, install it first:

pip install uv

Install:

uv tool install --python 3.12 ibm_watsonx_data_integration_mcp

Verify installation:

data-intg-mcp --help

MCP server configuration:

{
  "mcpServers": {
    "data-intg-mcp": {
      "command": "uv",
      "args": [
        "tool",
        "run",
        "data-intg-mcp"
      ],
      "env": {
        "WATSONX_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

Option 3 — Using pip with a Virtual Environment

Install:

python3.12 -m venv .venv
source .venv/bin/activate
pip install ibm_watsonx_data_integration_mcp

Verify installation:

data-intg-mcp --help

MCP server configuration:

{
  "mcpServers": {
    "data-intg-mcp": {
      "command": "<Path to .venv>/bin/data-intg-mcp",
      "args": [],
      "env": {
        "WATSONX_API_KEY": "YOUR_API_KEY"
      }
    }
  }
}

AI Agent Client Integration

Claude Desktop — Navigate to Claude, select Settings → Developer, and select Edit Config. Add one of the MCP server configurations above to your claude_desktop_config.json file.

Bob (VS Code) — Open VS Code, open Bob, select Settings → MCP, and select Open Global MCPs. Add one of the MCP server configurations above to your Bob MCP config file.

Authentication

Set your WATSONX_API_KEY in the env block of your MCP client configuration. The server validates this key on startup.

Try it free

Create a free trial of watsonx.data integration and try the generally available MCP server today.