Overview:
LLMs are moving from experimentation to production, and enterprises are building AI agents that need access to real-time business data. watsonx Orchestrate offers conversational AI for workflow automation, but connecting it to enterprise systems like IBM Content Aware Storage (CAS) traditionally requires custom integration code for each data source.
The Integration Challenge:
- Every vector database integration means writing custom API client code
- AI agents can’t discover new capabilities without redeployment
- Authentication, error handling, and connection management add development overhead
- Scaling to multiple data sources multiplies the complexity
Model Context Protocol (MCP) fixes this by providing a standardized way for AI agents to discover and interact with enterprise systems conversationally—no custom integration code needed.
What this article covers: This article explains how MCP enables watsonx Orchestrate to access IBM Content Aware Storage, demonstrates the technical implementation, and shows the integration in action.
What is Model Context Protocol (MCP)?
MCP is an open protocol (JSON-RPC 2.0) that enables AI agents to discover and use enterprise tools through natural language. Instead of building custom integrations for every system, MCP provides a standardized interface where AI agents query available capabilities at runtime and automatically call the appropriate backend services based on user requests.
Why MCP Matters for Enterprise AI ?
Traditional AI models can't access your latest documents, archived records, or real-time business information. MCP solves this by letting AI agents query live enterprise systems during conversations, ensuring responses are grounded in verified data rather than outdated knowledge. It eliminates custom API development, enables dynamic capability discovery, and enforces security controls at the data source level.
IBM Fusion HCI and Content-Aware Storage
IBM Fusion HCI is a hyperconverged infrastructure platform purpose-built for AI and analytics workloads. It unifies compute, storage, and networking into a single system optimized for GPU-accelerated applications and low-latency data access.
Content-Aware Storage (CAS) is the enterprise vector database, designed specifically for Retrieval-Augmented Generation (RAG) applications.
Some of CAS capabilities include:
- Vector embeddings storage: Stores and indexes document embeddings at enterprise scale
- Semantic search: Performs similarity searches across millions of vectors
- REST APIs: Full APIs for vector store management, ingestion, and search operations
The challenge: CAS provided powerful REST APIs, but accessing them from AI agents still required custom code — authentication handlers, connection management etc. AI agents couldn’t dynamically discover what vector stores were available or adapt to new capabilities without code changes.
The Solution: MCP for CAS
We implemented MCP as a protocol layer on top of CAS, enabling AI agents to interact with vector stores conversationally.
1. Standard endpoint: /cas/api/v1/mcp
- Synchronous JSON-RPC 2.0 requests
- Best for: Single queries, simple workflows
2. Streaming endpoint: /cas/api/v1/mcp-streamable
- Streamable http transport for real-time responses
- Best for: Streaming Clients
Authentication: Uses existing CAS bearer tokens — same security model, zero additional complexity.
Proven Test & Integration
With the architecture in place, the next step is demonstrating how MCP eliminates custom integration code in practice. This section walks through WatsonX Orchestrate for validation.
The integration follows three key phases: Connecting CAS to watsonx Orchestrate via MCP, creating an AI agent with CAS tools, and validating queries through natural language interaction.
Prerequisites
Before starting, ensure the following are in place:
- Watsonx Orchestrate instance — An active Orchestrate deployment with CLI access. If you haven’t set this up yet, follow this official installation guide.
- Add a model to your Orchestrate Deployment — To run AI agents that query CAS, make sure your instance has a model configured. Follow these steps to add model.
- IBM Fusion instance with CAS deployed — CAS instance running on OpenShift. Visit for installation steps.
- CAS authentication token — Bearer token for API access.
- CAS route URL — OpenShift route where CAS is accessible , for example:
https://<console-ibm-spectrum-fusion-ns.apps.openshifturl>
- Network connectivity — The watsonx Orchestrate instance must have network access to the CAS server route.
Configuration Steps:
Step 1: Set Up the watsonx Orchestrate CLI Environment
The watsonx Orchestrate CLI allows you to manage MCP servers, toolkits, and agents directly from your terminal.
1.1 Install the watsonx Orchestrate CLI
Download and install the CLI for your operating system (macOS, Linux, or Windows)
watsonx Orchestrate CLI installation guide: https://developer.watson-orchestrate.ibm.com/getting_started/installing
Verify installation: orchestrate --version
1.2 Set up Python virtual environment (recommended):
python3 -m venv orchestrate-env
source orchestrate-env/bin/activate
You’ll need watsonx Orchestrate instance URL & its credentials (username/password) later.
Step 2: Configure Orchestrate Environment
Get your Orchestrate instance URL from Profile Settings in the Orchestrate Dashboard.
orchestrate env add -n wxo-instance -u <your-orchestrate-url>
# Activate the environment
orchestrate env activate wxo-instance**Note** You'll be prompted for credentials—enter your Orchestrate username and password.
Step 3: Prepare the CAS Server URL
Retrieve the CAS route from your OpenShift dashboard. To help you locate the route, here’s a screenshot of the OpenShift dashboard showing the CAS route:
Add “/cas/api/v1/mcp-streamable/” in end of the Route. Example — “https://<cas-route>/cas/api/v1/mcp-streamable/”
Step 4: Register CAS MCP Server
Add the CAS MCP server as a toolkit in watsonx Orchestrate:
orchestrate toolkits add \
This command registers the CAS MCP server and makes both tools available to Orchestrate agents.
Step 5: Create and Configure Agent
- Access Orchestrate UI
- Navigate to your watsonx Orchestrate dashboard
- Choose tools from “cas-vector-search” toolkit
- Select both
list_vector_stores and search_vector_stores
- Choose tools from “cas-vector-search” toolkitClick “Create Agent”
2. Add MCP Tools
- In agent configuration, select “Add Tools”
- Choose tools from “cas-vector-search” toolkit.
- Select both list_vector_stores and search_vector_stores
3. Save Configuration & Deploy Agent
Validation
— Tool Listing after registering MCP Server
— Verify Tool Execution
Once deployed, test the agent with natural language queries:
Example prompts:
“What vector stores are available in CAS?”
Provide URL and token as prompted by agent in chat.
What We Achieved:
We’ve successfully connected IBM CAS to watsonx Orchestrate, enabling AI agents to query enterprise vector stores directly through natural language. With this integration, you can:
- Run agentic workflows — Let Orchestrate autonomously retrieve and act on enterprise knowledge.
- Leverage RAG pipelines — Seamlessly connect MCP-compatible AI agents to CAS vector stores.
- Avoid custom code — Orchestrate handles the connection and workflow orchestration for you.
Explore further:
- For details on IBM Fusion, please refer our product page.
- For more protocol details and implementation examples, visit our official page.
- Learn more about MCP and CAS architecture here.
- An end-to-end demo of this integration is available as a YouTube video here.