Exploring the Potential of the Model Context Protocol (MCP) to Extend Oracle EPM into AI-Driven Workflows.

Introduction

In recent months, discussions about the integration of AI with enterprise systems have gained substantial momentum. Financial services, in particular, are investigating how Large Language Models (LLMs) and Generative AI could enhance existing platforms such as Oracle EPM, which remains the cornerstone of scenario planning and financial forecasting in most banks.

One emerging standard that could influence these discussions is Anthropic’s Model Context Protocol (MCP). Designed as an open protocol to connect AI agents with external systems, MCP promises a more consistent and secure way to extend AI-driven workflows into enterprise landscapes.

In this article, we explore how MCP might interface with Oracle EPM within the context of a finance transformation initiative.

Disclosure: This blog was created based on a conversation (or series of prompts) with Revvy, a ChatGPT developed by Revvence and trained on finance transformation and systems change-related content.

Jump to Section

1. Understanding Model Context Protocol (MCP).
2. How EPM and MCP Would Communicate.
3. Teams Integration and Human-in-the-Loop.
4. Looking Ahead: The Art of the Possible.
5. Conclusion.
6. How can we help?

Oracle EPM's Strengths and Boundaries

Oracle EPM is, at its core, a powerful, in-memory platform for planning, scenario modelling, and financial consolidation. Its driver-based models, scenario capabilities, and governance structures make it well-suited for supporting board-level decision-making, especially in banking, where macroeconomic variables (MEVs) can impact balance sheets and profitability.

However, EPM’s Narrative Reporting capabilities, while useful for structured disclosures, often lack the flexibility and interactivity that modern finance teams increasingly expect. Teams often find themselves relying on separate tools, such as Workiva, for final-mile reporting.

Simultaneously, the rapid advancement of generative AI, particularly with tools like ChatGPT, has highlighted the potential for more dynamic and interactive narrative content creation. Finance teams are already experimenting with generative AI to draft management commentaries, produce summaries, and even automate first-pass regulatory reports.

However, these capabilities often function in isolation from trusted data sources, raising concerns regarding accuracy, governance, and auditability.

This is where Oracle EPM’s structured, governed data can play a critical role. By integrating EPM’s scenario outputs with generative AI capabilities, finance teams can benefit from the best of both worlds: dynamic, conversational narratives grounded in authoritative, scenario-driven data. This approach would enhance EPM’s narrative reporting capabilities, enabling finance teams to collaborate and operate more efficiently while maintaining the trust and governance required in regulated environments.

Understanding Model Context Protocol (MCP)

Model Content Protocol (MCP), introduced by Anthropic in late 2024, is an open protocol designed to standardise how AI agents (such as large language models) connect with external systems in a secure, consistent, and context-aware manner. Think of it as a universal connector—sometimes called the “USB-C of AI integration”—that enables a single AI agent to talk to multiple systems in a uniform way.

In practical terms, MCP defines:

• A common language and structure for interactions between AI assistants and enterprise systems
• How they exchange data.
• Authenticate requests.
• Handle errors.
• Manage governance.

Rather than requiring every AI application to create its own unique connector for each system—an approach that rapidly becomes unmanageable in extensive finance architectures—MCP provides a standardised method that simplifies integrations while maintaining control and compliance.

In the context of a finance systems landscape, this has significant implications. Financial institutions typically operate a complex web of systems, including planning and forecasting platforms such as Oracle EPM, ERP systems, regulatory reporting engines, risk models, data warehouses, and collaboration tools like Microsoft Teams. Integrating AI with enterprise platforms can be a complex and expensive process.

MCP starts to reduce the cost and complexity by enabling an AI agent to:

• Request data from systems like Oracle EPM using a consistent request/response structure.
• Execute actions (e.g. running a scenario calculation, triggering a forecast refresh) using pre-defined, secure interfaces.
• Handle authentication in a standardised way—often using OAuth 2.0 or enterprise identity platforms.
• Log and audit interactions to ensure regulatory compliance and traceability.

By decoupling the AI agent from the specifics of each system’s API design, MCP enables banks to preserve their existing system architectures (such as EPM’s REST APIs and Groovy jobs) while allowing AI agents to interact with those systems more flexibly.

For instance, an AI assistant might request the latest scenario forecasts from Oracle EPM via MCP, then generate a draft narrative for the finance team in Microsoft Teams—all without needing to build a bespoke integration for each step.

This model is particularly powerful in finance because it aligns well with governance and risk management requirements. MCP’s structured approach to authentication, role-based access, and logging ensures that AI integrations can be monitored and controlled, which is essential in an industry where regulatory scrutiny is high.

In summary, MCP’s value proposition in the finance systems landscape is that it enables banks to explore AI-driven workflows, such as narrative reporting and regulatory disclosures, while preserving the integrity of the underlying platforms that finance teams rely on.

MCP represents a practical and strategic step towards creating a more connected, agentic finance function, integrating trusted data sources like Oracle EPM with the power of generative AI and AI Agents to enhance collaboration and boost productivity.

How EPM and MCP Would Communicate.

For MCP to function effectively with Oracle EPM, a technical integration layer is necessary—usually implemented as a dedicated MCP server or middleware application. This layer would serve as an MCP-compliant server, responding to AI agents’ requests while handling the specifics of interfacing with EPM’s REST APIs and Groovy scripting interfaces.

Oracle EPM itself does not natively speak MCP. Instead, it exposes data through its REST API, which supports operations such as:

• Exporting data slices from Planning cubes (using /applications/{appName}/plantypes/{planType}/exportdataslice).
• Triggering business rules or Groovy jobs that can calculate scenario outputs or refresh data.
• Managing metadata and other administrative tasks.

An MCP server, sitting between the AI agent and EPM, would need to:

• Authenticate securely to EPM (using OAuth 2.0 or Basic Auth, depending on the implementation).
• Fetch scenario data from EPM via REST API calls or by executing a Groovy script that generates tailored outputs (e.g. forecasts, cost data).
• Transform the EPM data (which is typically hierarchical and multi-dimensional) into a structured format (e.g. JSON) that an AI agent can understand.
• Handle batching and pagination, since EPM’s API responses may need chunking for large data slices.
• Enforce governance, ensuring that only pre-approved data slices or scenario outputs are exposed to the AI agent—never raw, unrestricted cube access.

Once the MCP server has gathered and transformed the data, it responds to the AI agent using the MCP protocol, thereby allowing the AI assistant to incorporate the data into its reasoning chain, generate narratives, or respond to finance team queries.

Supporting Technology Platforms.

To make this architecture viable, a few additional components are typically required:

• API Gateway (e.g. Azure API Management, Kong, Apigee): To secure and manage incoming and outgoing API calls between MCP, EPM, and other enterprise systems. In this scenario, we assume the bank is predominantly a Microsoft shop; therefore, Azure API Management would be a natural fit. This layer can handle authentication, rate limiting, and logging.
• Enterprise Identity Management (e.g. Azure Active Directory, Active Directory Federation Services): To ensure that role-based access controls govern AI agent interactions and that requests to EPM carry appropriate credentials. Again, assuming a Microsoft-centric environment, Azure Active Directory would typically handle OAuth 2.0 and SAML authentication flows, enforcing secure connections between MCP, EPM, and the AI assistant.
• Containerised Middleware (e.g. Kubernetes or Docker Swarm): To host the MCP server itself. Banks often choose container orchestration to manage scalability, failover, and patching.
• Data Transformation Layer (e.g. dbt, Apache NiFi, or custom Python modules): To convert EPM’s multi-dimensional outputs into flat structures that are easier for AI models to interpret.
• Microsoft Teams Integration Layer, typically implemented using the Microsoft Graph API, to allow AI-generated narratives to be delivered directly to finance teams within Teams. This fosters collaboration and review workflows without forcing users to switch platforms.

Identity Management & Security Considerations
In many banks, Active Directory or Azure Active Directory forms the backbone of identity management, providing authentication and authorisation for users and systems. This directory service is typically extended with OAuth 2.0, SAML, and role-based access controls to govern secure access to APIs and data. In the MCP-EPM integration scenario, we assume a Microsoft-centric environment where Azure Active Directory would handle authentication tokens for both the API Gateway and the MCP server itself, ensuring that enterprise security policies govern all interactions with EPM.

Furthermore, security and governance cannot be an afterthought. Any architecture that bridges EPM with AI assistants must ensure robust authentication, role-based access controls, and audit trails. MCP provides the framework for this, but banks will need to design their implementation carefully to satisfy regulatory requirements.

Teams Integration and Human-in-the-Loop.

For finance teams already utilising Microsoft Teams as their primary collaboration platform, integrating MCP into Teams could offer a more seamless experience. Draft narratives or scenario summaries generated by the AI assistant could be delivered directly into Teams channels or chats, enabling finance professionals to review, edit, and approve content within their existing workflows.

This human-in-the-loop approach is essential; while generative AI is powerful, it cannot produce final regulatory disclosures without human oversight.

Furthermore, MCP’s ability to manage authentication and logging ensures that all interactions between the AI agent and EPM data are appropriately audited. This is particularly essential in banking, where regulators are increasingly requiring evidence of data lineage and control over how data-driven insights are generated.

Looking Ahead: The Art of the Possible.

While the current focus of MCP may be on enabling narrative reporting and scenario summaries, its true potential lies in unlocking entirely new ways of working in finance. As banks continue to integrate generative AI capabilities with trusted data sources such as Oracle EPM, MCP could be the key to orchestrating more agentic workflows that automate—and even enhance—traditional finance tasks.

One of the most immediate areas where MCP could add value is in narrative reporting. Traditionally, narrative reporting involves structured, repeatable commentary that explains actual or forecasted financial results, such as management discussions and analysis, board packs, or regulatory disclosures.

An MCP-enabled AI assistant could generate these reports by drawing on EPM’s aggregated data, variance analyses, and standardised metrics, drafting commentary in the bank’s preferred tone and style, complete with charts and explanations. Finance teams would then focus on reviewing, refining, and approving the output rather than building it from scratch.

At the same time, MCP could facilitate dynamic scenario commentary, wherein AI agents engage with EPM’s scenario modelling outputs to generate forward-looking insights. For instance, once a scenario is finalised in EPM, an AI agent could explain how a 0.5% GDP growth scenario would affect profitability, capital ratios, or risk-weighted assets (RWA). This agent could provide additional context by referencing real-time market data or regulatory changes, ultimately creating more insightful scenario commentaries.

In a more advanced state, MCP could support fully orchestrated, end-to-end workflows, where AI agents integrate Oracle EPM’s anomaly detection (via Insights), variance analysis, and scenario modelling with external data sources to build comprehensive, management-ready report packs.

For instance, after a scenario is finalised in EPM, an MCP-enabled AI assistant could automatically consolidate key variances detected by EPM Insights and standard reports, cross-reference them with external economic indicators, and use a Retrieval-Augmented Generation (RAG) approach to draft a complete narrative in the bank’s preferred tone and style. This AI-generated report could incorporate relevant regulatory references, commentary on capital impacts, and recommended management actions, such as rebalancing the cost base or other strategies.

By enabling the AI agent to integrate context from both EPM and external sources, MCP allows finance teams to move beyond static, isolated reports and towards dynamic, narrative-rich report packs that encompass the full complexity of the bank’s environment. This not only enhances decision-making but also releases finance professionals to concentrate on value-added analysis rather than manually assembling data from multiple systems.

Conclusion.

The Model Context Protocol presents a promising approach to expanding the capabilities of Oracle EPM into more comprehensive AI-driven workflows, particularly in areas such as narrative reporting and dynamic scenario commentary. While real-time conversational querying remains technically challenging (read our blog "Talking to Your Data: Rethinking Internal Reporting" for more on this topic), MCP bridges the gap between EPM’s structured scenario outputs and the dynamic, interactive experiences that finance teams increasingly expect.

As banks continue to innovate, MCP stands out as a foundational building block for enabling more agentic finance processes, where AI assistants collaborate seamlessly with trusted data to generate insights, propose scenario-driven recommendations, and enhance decision-making. By integrating generative AI with EPM’s core strengths, MCP unlocks new possibilities for finance teams to devote more time to strategic analysis and business partnering, while reducing the need for manual tasks.

How can we help?

Revvence can help in several valuable ways:

• Check out Revvy, our Narrow-GPT for Finance Transformation. Read all about Revvy here.
• Review one of your existing finance processes to recommend where AI capabilities will have the most impact.
• Conduct one of our Innovation Labs (a free three-hour workshop) to show you the art of the possible and help you build your business case for change.
• The design and delivery of end-to-end solutions.