Creating MCP tools

Edit online

The Model Context Protocol (MCP) provides standardization that allows APIs to be represented as tools that AI agents can discover and invoke.

Enterprise capabilities such as retrieving data or updating records are typically exposed through REST APIs. These APIs and its operations are commonly defined by using OpenAPI specifications. However, AI agents or large language models (LLMs) cannot understand the OpenAPI specifications. For an AI agent to understand what actions are available and how to invoke them, these capabilities must be exposed in a standardized MCP format.

An MCP tool represents a specific action that an AI agent can perform, such as retrieving or updating data. Each MCP tool corresponds to an API operation.

An MCP server hosts one or more MCP tools and exposes them in a format that AI agents can discover and invoke. The MCP server acts as the interface through which AI agents interact with enterprise capabilities.

In IBM API Studio, you can create MCP tools in the following ways.
  • Convert existing OpenAPI based APIs into MCP tool definitions and generate an MCP server.
  • Register an external (third-party) MCP server and use its MCP tools.
After you create or register MCP tools, you can apply governance policies and publish the MCP server so that AI agents can securely discover and invoke these capabilities.

Benefits of converting APIs into MCP tools by using IBM API Studio
  • Select relevant API operations.

    Choose only the API operations that you want AI agents to access so that exposed capabilities are aligned with your use case.

  • Simplify MCP tool generation.

    Automatically generate MCP‑compliant tools from existing OpenAPI definitions without modifying your underlying APIs.

  • Improve tool descriptions for AI agents.

    Use the MCP server enhancer to refine tool names and descriptions so that AI agents can understand how and when to use each tool. It refines OpenAPI technical field descriptions into structured LLM-optimized metadata that helps AI agents understand the semantic context of each tool's operations, parameters, prerequisites, and possible outcomes.

  • Improve invocation reliability.

    Clearer tool definitions help AI agents select correct tools and parameters. It reduces the hallucination gaps, resulting in fewer failed API calls during autonomous task execution.

IBM API Studio offers flexible options for creating and updating the MCP servers.

  • Form view. Create and manage MCP server using a structured form-based UI.
  • Code view. Update MCP resources in YAML format within the code editor for direct control over the MCP definitions. This approach works well for those who prefer working with code.

Alternatively, you can use Bring your own coding agent (BYOCA) functionality to access DataPower Interact Gateway (IDIG) capabilities from Model Context Protocol (MCP) clients such as Bob, VS Code extensions, and Claude Desktop. For details, see Bring your own coding assistant (BYOCA)