Interacting with data through an MCP server

The watsonx.data MCP server is currently available only for instance administrators. Users with non-admin roles (such as USER role) cannot access MCP tools or use the watsonx.data assistant's MCP functionality, even if they have been granted access to engines.

You can securely access and explore your lakehouse data and metadata through natural language by using watsonx.data Model Context Protocol (MCP) server and your AI agent.

The IBM watsonx.data MCP Server is designed for the following users:

  • AI agent developers who are building data-aware assistants
  • Platform teams who are enabling governed AI access to lakehouse data
  • Users who want conversational querying without exposing write access

Connection models

You can choose between a remote or a local MCP server. Both types provide the same tools and capabilities.

Remote MCP server
A fully managed endpoint that is hosted on IBM Cloud.
Local MCP server
A locally installed server that runs on your local system.

The following table compares the characteristics of the remote and local MCP server implementations.

Characteristics of remote and local MCP servers
Characteristic Remote MCP server Local MCP server
Installation and resource requirements None. Maintained by IBM. Local installation and resource access required.
Intended use Managed, scalable, and governed access Highly customizable, environment-specific
Connectivity environment Internet connection required. Internet connection required.
Supported hosts Any MCP-compliant host Any MCP-compliant host
Integration and customization Standardized, platform-defined integration Fully customizable, user-defined integration

Capabilities

Both remote and local MCP servers provide the same capabilities. The IBM watsonx.data MCP Server provides the following capabilities:

Query execution
Execute SQL queries against your lakehouse data by using natural language.
Catalog operations
Browse, search, and retrieve metadata from your data catalog.
Data ingestion
Load and manage data into your watsonx.data instance.
Engine management
Monitor and interact with your query engines.
Data security
Secure your data through IBM Cloud Identity and Access Management (IAM) authentication with automatic token refresh. Write access is controlled to allow INSERT and UPDATE operations while restricting DELETE operations.

For the complete list of available tools and detailed use cases, refer https://github.com/IBM/ibm-watsonxdata-mcp-server/blob/main/TOOLS.md.

AI agent integrations

You can use the IBM watsonx.data MCP server with the following widely adopted MCP hosts:

  • IBM Bob
  • Claude Desktop
  • Other MCP-compatible clients. Any client that supports the MCP protocol can connect to the remote server

MCP server architecture

The following diagram illustrates the process of querying data through the remote MCP server:

architecture diagram
Remote MCP architecture diagram

The following diagram illustrates the process of querying data through the local MCP server.

architecture diagram
Local MCP architecture diagram

The MCP server enables conversational data access through the following workflow:

  1. You submit a natural language query to your agent.
  2. The agent interprets your request and determines the appropriate action.
  3. The agent communicates with the MCP server, which then forwards the request to your watsonx.data instance.
  4. watsonx.data processes the request and returns results to the MCP server.
  5. The agent presents the results in a conversational format.

Next steps