Integrating Fusion Application AI Agent Studio with OIC as an External System

Introduction

When building agents in Fusion Application AI Agent Studio, integration with external systems is often essential. Many Fusion Application customers already use Oracle Integration Cloud (OIC) to connect Fusion applications with external services. The same Oracle Integration Cloud integrations can now be exposed as tools for AI agents, enabling agents to interact with external systems in a more dynamic and scalable way.

In this blog, we will explore the different ways Oracle Integration Cloud can expose integrations as tools for AI agents, along with the options available in Fusion AI Agent Studio for calling OIC integrations.

Integration as Tools

Oracle Integration Cloud provides two options for exposing integrations as tools for an AI agent.

An integration can be created using the REST Adapter with the Trigger option enabled on the connection. Such integrations can be configured directly as External REST tools in Fusion AI Agent Studio, with no additional setup required.

Direct Integration

Integrations created both inside and outside an OIC project can be used as tools. They can be monitored through the standard observability layer, and when created within a project, they can also be monitored at the project level.

MCP Server

MCP is an industry-standard approach for tool exposure and allows the AI agent to discover tools dynamically at runtime. The MCP Server option is available only within OIC projects. An OIC project provides a unified workspace where teams can design, manage, and monitor automation solutions, including integrations, AI agents, B2B, decisions, healthcare, human-in-the-loop workflows, and robotic automation. It also simplifies lifecycle management and streamlines updates for prebuilt integrations.

To configure MCP server, you first create a project in OIC and then add the required integrations inside that project. To expose these integrations as tools through MCP, add a tool in the AI Agents section of the OIC project and associate it with the relevant integration. A meaningful tool description and parameter metadata are important, as they help the AI agent identify the correct tool based on the user’s request. It is not necessary to create tools for every integration in the project. You can expose only the integrations intended for AI agent use.

Once the tools are defined, the project can be enabled as an MCP server. All tools within that project become part of the MCP endpoint, which can then be configured in Fusion Application AI Agent Studio as an MCP tool. OIC MCP does not introduce additional observability. The underlying integrations continue to be monitored through the standard observability layer, both inside and outside the project.

The following image illustrates a sample ecommerce project with three integrations for retrieving order, shipping, and invoice details, along with the corresponding three tools included in the Ecommerce MCP Server.

External Integration Tools for Fusion Application AI Agents

Fusion AI Agent Studio provides various options for configuring tools that connect to external systems. Both OIC integrations exposed directly and integrations exposed through MCP can be configured as tools, depending on the capabilities supported by Fusion AI Agent Studio.

Direct Integration

There are two ways to configure direct access to OIC integrations in AI Agent Studio

1. External REST

In this approach, the integration is configured as an External REST tool type. This option provides multiple authentication methods for connecting to external integrations. For OIC integrations, authentication can be handled through either:

• OAuth 2.0 Client Credentials
• OAuth 2.0 Resource Owner Password Credentials

Both options require a confidential application in the OIC Identity domain. The OAuth 2.0 Client Credentials flow additionally requires the confidential application to have the Service Invoker role. With the client credentials flow, the integration is invoked through the application context. With the resource owner password credentials flow, the integration is invoked using a shared integration user configured in the authentication settings of the External REST tool. Neither of these methods supports identity propagation for the logged-in user. For OIC Integration, OAuth 2.0 Client Credentials is generally the preferred option.

{
    "grant_type":"client_credentials",
    "client_id":"xxxxxxxxxxxxxxxxxxxxxx",
    "client_secret":"xxxxxxxxxxxxx",
    "scope":"https://xxxxxxxx.integration.us-region.ocp.oraclecloud.com:443urn:opc:xxxxxxxx::all https://xxxxxx.integration.region.ocp.oraclecloud.com:443/ic/api/"
}

2. Business Object:

In this approach, the integration is configured as a business object and then wrapped in a tool of type Business Object. The business object supports multiple resource type options, and the integration is configured using the Other Data Source Application resource type. This option requires a Data Source Application configuration that includes the Token URL, Client ID, Scope, Private Key, and Public Key. It generates an access token using a user assertion based on the logged-in user, which enables identity propagation between Fusion Application and Oracle Integration Cloud. Please refer to the identity propagation blog for a detailed implementation guide.

MCP Server

The OIC MCP server can be configured as an MCP tool in Fusion AI Agent Studio. This tool type supports both SSE and Streamable HTTP options. For OIC MCP, the Streamable HTTP option is recommended.

This tool type also supports multiple authentication methods. OIC MCP can be configured using Client Credentials by providing the relevant configuration details in the JSON payload. Fusion AI Agent Studio also supports JWT authentication, but this approach is tied to a single user through the sub claim and requires private key and claim configuration. For OIC MCP server scenarios, Client Credentials is generally the preferred option.

It is important to note that OIC MCP configuration in AI Agent Studio does not support identity propagation. Fusion AI Agent Studio also allows you to select only specific tools from an MCP server. This is useful when an agent needs only a subset of the available tools. For example, if an MCP server exposes 10 tools but the agent needs only 3, this selection helps optimize token usage because only the chosen tool definitions are passed to the agent.

The following image shows the AI Agent Studio MCP tool, which allows you to select the tools exposed by the OIC MCP server.

Fusion Application AI Agents

Once tools have been configured, they can be used within Fusion Application AI Agents. Fusion AI Agent Studio provides different ways to consume tools exposed through OIC direct integrations and the OIC MCP server.

Direct Integration

Direct integration tools configured as External REST or Business Object can be used in both Agent and Workflow scenarios in Fusion Application 26C.

• Agent: The agent can be configured with either External REST or Business Object tools. When a Business Object tool is used, the system generates the access token using the logged-in user’s identity and invokes the OIC integration accordingly.
• Workflow Agent: OIC integrations can be invoked directly through the External REST node or the Business Object Function node. When configured as a Business Object Function node, the system again uses the logged-in user’s identity to generate the access token and call the OIC integration.

MCP Server

The OIC MCP Server tool can be used only within an Agent in Fusion Application 26C. It cannot be invoked directly from a workflow. Workflows can access it indirectly through an Agent node. Unlike direct integration, the MCP Server invocation does not use the logged-in user’s identity. It is authenticated through the client credentials flow.

Summary

MCP is an industry-standard approach for exposing tools to an agent, and it can be enabled in Oracle Integration Cloud with a single click. Fusion AI Agent Studio allows you to select only the tools you want from an MCP server, ensuring that only the required tools are included in the agent’s context window.

Direct integration is best suited for scenarios where identity propagation from Fusion Application to Oracle Integration Cloud is required, or when integration needs to be called directly from an agent or workflow. MCP is a stronger fit when an agent needs to dynamically decide which integration to invoke based on the user’s request. Choose MCP Server when leveraging the industry-standard MCP protocol and enabling dynamic tool discovery at runtime.

Fusion AI Agent Studio: Direct Integration vs MCP Server

Conclusion

Fusion Application AI Agent Studio, together with Oracle Integration Cloud, provides a flexible way to extend AI agents with enterprise integrations. Whether you choose Direct Integration for simpler connectivity and identity propagation, or MCP Server for standardized, dynamic tool discovery, both approaches enable agents to interact with external systems in a secure and scalable manner. The right choice depends on your use case, Direct Integration is ideal when user context and workflow invocation are important, while MCP Server is better suited for scenarios where agents need to select from multiple tools at runtime.