Introduction

Over the last year, AI has become highly effective at understanding language and generating content. However, in enterprise environments, the real challenge is not generation—it's accessing and acting on business data.

Take invoice generation as an example. Creating an invoice requires customer, product, order and pricing information, often spread across multiple systems. Without access to this enterprise context, AI cannot produce accurate and business-ready results.

This is where IBM App Connect MCP helps. By exposing existing APIs and integrations as MCP tools, it enables AI agents to securely discover and invoke enterprise capabilities without changing existing business logic. To demonstrate this, I built an Invoice Automation solution that allows an AI agent to generate enterprise-ready invoices by interacting with enterprise systems and databases through MCP.

Solution Overview

→ Rather than embedding business logic inside the AI Agent, all execution responsibilities remain within App Connect Enterprise.

The AI Agent determines → User intent, Required actions and Tool invocation sequence

App Connect handles → Data retrieval, Data enrichment, Validation, Persistence and Integration with enterprise systems 

This separation ensures that enterprise governance remains within existing integration layers while still enabling AI-driven user experiences.

Solution Architecture

The solution architecture consists of three major components.

1. AI Agent : The AI Agent acts as the orchestration layer. When a user submits a request such as: “Generate an invoice for order 1001.” the Agent determines which MCP tools are required and orchestrates the overall workflow.

2. App Connect MCP Layer : App Connect Enterprise MCP exposes REST API operations as discoverable tools that can be invoked by an AI agent. This capability allows existing integrations to become AI-accessible without requiring application changes.

3. SKILL.md : A dedicated skill file defines invoice generation standards including required fields, Invoice structure, formatting guidelines, output expectations and enterprise data.

REST APIs

Keeping business logic within App Connect provides several advantages:

• Existing governance controls remain intact

• APIs can be reused beyond AI use cases

• Enterprise validation rules remain centralized

• Integrations stay independent of any specific AI platform

The project was structured into multiple message flows, each corresponding to a distinct stage of the invoice lifecycle.

newOrder Creates an order and enriches it with customer and product information. newProductNCustomer Allows new customer and product information to be added to the system. saveOrder Persists validated order information. getOrder Retrieves existing order data for invoice generation. saveInvoice Stores finalized invoices for future access and auditing.

Exposing APIs as MCP Tools

Once the APIs were deployed, the next step was exposing them through an MCP server. Starting with IBM App Connect Enterprise 13.0.7.0, deployed REST API operations can be surfaced as MCP tools that AI agents can discover and invoke.

This allows existing integrations to be made AI-accessible without changing the underlying business logic. After publishing the MCP server, the APIs became available as MCP tools, enabling the AI Agent to dynamically discover and orchestrate invoice automation capabilities through a standardized MCP interface.

Connecting the AI Agent

For this demonstration, I used IBM Bob v2 as the AI Agent. Bob connects to the App Connect MCP server over a Streamable HTTP endpoint (/mcp), allowing it to communicate with the MCP server and discover available tools. Once connected, the Agent automatically retrieves the list of exposed capabilities and their descriptions.

Notice how the Agent is not hardcoded to specific APIs; instead, it dynamically discovers and uses tools through MCP. This abstraction separates tool implementation from AI reasoning, while MCP settings allow tools to be enabled or disabled as needed.

Prompting

With the MCP connection established, I tested a simple request. listing the tools available on our MCP server.

By querying the MCP server, the AI agent automatically identifies all the invoice automation capabilities exposed through App Connect Enterprise, along with their descriptions and expected inputs. As new integrations are exposed through MCP, they become immediately available to the agent, making the solution both scalable and easy to extend.

Next we can prompt it with something like

generate an invoice for order id 10000

The Agent performs several actions behind the scenes. It begins by identifying the user's intent and determining which MCP tools are required to complete the request. Before invoking MCP tools or executing operations, the Agent seeks user approval, providing an important layer of governance and transparency. This ensures that users remain in control of what actions are performed against enterprise systems, particularly when write operations are involved. Similarly, before using the invoice-generation skill, the Agent requests approval, allowing users to verify exactly what instructions are being applied.

Before executing any create, update or delete operation, the Agent performs an intent confirmation step to ensure that the requested action is understood correctly. As shown above, rather than immediately creating the order, the Agent summarizes the customer and product details it has interpreted from the prompt and asks for explicit user confirmation before proceeding. This introduces an additional layer of control and governance, helping prevent accidental changes to enterprise data while ensuring that business transactions are executed exactly as intended. In enterprise environments, this human-in-the-loop approach is critical for maintaining accuracy, trust and auditability.

Once approved, the Agent invokes the GetInvoiceByOrderId tool to retrieve the required order and invoice data from backend systems exposed through App Connect MCP. It then applies the predefined invoice-generation instructions from the skill file, ensuring that the output follows a consistent, enterprise-approved structure. Using the retrieved business data and approved rules, the Agent generates a structured invoice as an HTML artifact, which can also be exported as a PNG image for sharing or storage. This demonstrates how AI agents can go beyond content generation and securely orchestrate business workflows while maintaining the control, validation and oversight required in enterprise environments.

For a prompt like this:

create a new order for customer 1002, who has bought 1 P112, 2 P118, 1 P105. give discount of 4 percent

One aspect of the solution I found particularly interesting is that the Agent does more than retrieve data. It can execute business operations. To demonstrate this, I asked the Agent to create a new order.

Behind the scenes, the Agent orchestrates multiple tool invocations. Possible workflow: newProductNCustomer2→ newOrder3→ saveOrder and even iterations of these.

If validation issues occur, the Agent requests the required information and continues the workflow accordingly. This demonstrates a key shift in enterprise AI. The Agent is no longer acting solely as a conversational interface. It is actively participating in business processes through governed enterprise integrations.

I have played with more prompts in the video here.

Why Skills Matter in this Usecase

An interesting observation during this implementation was the role played by the skill file.

Could the AI Agent generate an invoice without a predefined skill? Absolutely.

Would it generate the same invoice every time? Probably not.

In enterprise environments, consistency matters as much as intelligence. Different invoice structures can create challenges for Compliance, Reporting, Customer communication and Downstream processing

To address this, the Agent references a dedicated skill file that defines the required invoice structure.

The skill acts as a business contract. Regardless of the order being processed, the generated invoice follows the same layout Structure, formatting and Required fields. This combination of AI reasoning and controlled output is what makes the solution suitable for enterprise use cases. 

Conclusion

This example is ultimately not a story about generating documents. It is a story about connecting AI agents to real business capabilities and enterprise data securely. By combining IBM App Connect Enterprise with MCP, existing integrations can be transformed into discoverable tools that AI agents can use to execute enterprise workflows. Rather than embedding business logic inside AI systems, organizations can continue leveraging their existing APIs, governance controls and integration assets while enabling new AI-driven experiences.

Resources

Read more about MCP in ACE here

Check SKILL.md File here