Querying data through agents by using the MCP server
The IBM watsonx.data Model Context Protocol (MCP) Server enables agents to interact with IBM watsonx.data lakehouse instances through natural language interfaces. You can use this server to securely access and explore your lakehouse data and metadata through the Model Context Protocol, with built-in read-only protection to ensure data integrity.
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
Capabilities
The IBM watsonx.data MCP Server provides the following capabilities:
Data access and exploration
- Execute SQL SELECT queries using natural language or direct SQL syntax
- Browse data catalogs and schemas
- Inspect table structures and metadata, including columns, data types, properties, partitioning, and primary keys
- Monitor engine status and availability
Data security
- Read-only access enforcement (SELECT queries only)
- IBM Cloud Identity and Access Management (IAM) authentication
- Automatic token refresh mechanism
- Query validation and safety checks before execution
- Potentially unsafe operations are blocked
Transport mechanisms
- stdio transport for local subprocess communication. For implementation guidelines and security best practices, refer MCP Transports Specification.
AI agent integrations
-
Integrates with the following AI agents on your local computer:
- IBM Bob
- Claude Desktop
Configuration workflow
To configure the MCP server, complete these main tasks:
- Install and configure the MCP server on your local computer. See Installing and configuring the MCP server for querying data.
- Configure your AI agent to work with the MCP server and connect to watsonx.data. See Configuring IBM Bob or Configuring Claude Desktop.
MCP server architecture
The following diagram illustrates the process of querying data through the MCP server.
Understanding the interaction model
The MCP server enables conversational data access through the following workflow:
- You submit a natural language query to your agent.
- The agent interprets your request and selects the appropriate MCP tool.
- The MCP server executes the operation against your watsonx.data instance.
- Results are returned to the agent.
- The agent presents the results in a conversational format.