AI for Finance

Connect Claude to Oracle EPM. Safely. Practically. With an open-source MCP server.

Q: Is the FMEPM MCP server for Oracle EPM open source?

Yes. The public v1 of the MCP server is available on GitHub at github.com/fmepm with an MIT license. It includes 7 tools covering core Oracle EPM Cloud operations.

Q: Is it safe to connect AI to Oracle EPM?

Safety depends on how the integration is designed. The FMEPM MCP server respects Oracle EPM's native security model, authenticates through standard Oracle identity, and only exposes the specific operations you configure. For enterprise deployments, enterprise-grade AI platforms such as Amazon Bedrock, Azure AI Foundry, and Google Vertex AI offer Zero Data Retention (ZDR) and data residency controls.

AI assistants are now good enough to help your finance team run Oracle EPM workflows. The question is how to connect them without compromising security, data residency, or the integrity of your planning model. That is what this service delivers.

Start the conversation → View on GitHub →

Credentials: Oracle ACE Pro · Author, open-source Oracle EPM MCP server · Oracle EPM Cloud certified · Toronto, Canada. Serving clients worldwide.

Open source

The Oracle EPM MCP Server. Open source, built and maintained by me.

MCP (Model Context Protocol) is the open standard for connecting AI assistants to external systems. I published an open-source MCP server for Oracle EPM Cloud in April 2026, under the MIT license. It runs on your own computer and starts in mock mode, so you can try it before it touches a pod. Your Oracle sign-in sits in a config file on your machine, and the server can do only what your Oracle user can do. It has 7 tools: checking the API version, listing applications, reading and updating substitution variables, running business rules, exporting data slices, and polling job status.

github.com/fmepm/oracle-epm-mcp-server →

FMEPM Connect, the hosted edition, is live in beta. Claude today. Try it free for thirty days on your own test environment through sales@fmepm.com. The open-source core is free and self-managed. FMEPM Connect is the hosted and supported edition. Adding FMEPM Connect to Claude takes one minute, with nothing to install: watch the setup.

Three ways to connect Claude to Oracle EPM.

Offer	What it is	Who it is for	The next step
The open-source server	Free. You run it on your own computer, under your own Oracle sign-in. MIT license.	One admin or analyst who wants to try Claude on a test pod this afternoon.	Set it up in 5 minutes →
FMEPM Connect	The product. Hosted, or in your own cloud account. Each person signs in as themselves. Risky writes ask for your confirmation first, with a preview of what will change. Beta. Try it free for thirty days through sales@fmepm.com.	A finance team that wants Claude on its pod without running a server, with IT in the loop.	See the product →
Services	Implementation and training: the setup in your environment, the tools your workflows need, the security review with IT, and the sessions for finance users and admins.	A team that wants it done with them, and the people trained.	Start the conversation →

AI for Finance

AI integration for Oracle EPM Cloud

Connect AI assistants to Oracle EPM through an MCP server. Custom tools designed for your finance workflows. 2 to 8 weeks typical timeline.

MCP Claude Oracle EPM REST API Groovy Python AWS Bedrock Azure AI Foundry

Until recently, connecting AI to enterprise finance systems meant choosing between a brittle custom build or sending sensitive data to a public AI service. Neither fit regulated finance environments. MCP changes that. It creates a clean, governable boundary between the AI model and your Oracle EPM data, respecting your existing security and keeping control where it belongs.

What this looks like in practice: a CFO types a question. The AI turns it into an EPM operation, such as a data slice, a business rule or a report. It runs that through the MCP server with proper authentication. The figures come from your Oracle EPM application, not from the AI's memory. No copy-paste. No spreadsheet drift. The AI can still misread a figure or get a calculation wrong, so check key numbers before you share them.

When this makes sense for your finance team

Common situation

Your analysts spend hours generating the same variance reports each month. The data is in Oracle EPM, but extracting and formatting it is manual.

Common situation

Executives ask ad-hoc questions that require pulling specific EPM data. The answer takes a day because someone has to build the query.

Common situation

Your team wants to experiment with AI but cannot send EPM data to public AI services due to data residency or regulatory requirements.

Common situation

You have an idea for an AI-powered FP&A workflow but need someone who understands both Oracle EPM internals and how MCP actually works.

From recent work

Published an open-source MCP server for Oracle EPM Cloud in April 2026, under the MIT license. Since then: the security field guide for AI agents in Oracle EPM, in two parts, and the July 2026 MCP specification change, with what to do about it.

What's included in an engagement

MCP server deployment

The Oracle EPM MCP server deployed in your environment, authenticated against your Oracle EPM identity, with the tools your team needs.

Custom tool design

Additional tools built for workflows specific to your organization. Close checklists, variance explainers, rolling forecast assistants, whatever fits.

AI platform integration

Connection to your AI platform of choice. Claude Desktop for individuals, or enterprise platforms like Amazon Bedrock, Azure AI Foundry, or Google Vertex AI for production.

Security and governance

Configuration that respects your Oracle EPM security model, audit logging for AI-initiated actions, and review workflows for sensitive operations.

Team training

Hands-on sessions for both finance users (how to prompt effectively) and admins (how to extend and maintain the MCP server).

Documentation

A written record of your deployment: tools available, how they connect, how to troubleshoot, and what to do as MCP itself evolves.

Typical timeline: 2 to 8 weeks depending on custom tool scope and security review requirements.

Engagement model

How an engagement works

Four phases, consistent across every engagement. Senior-led from discovery through delivery.

Discovery

A working session to understand your Oracle EPM environment, the workflows you want to augment, and your AI platform constraints.

Architecture

Security model design, tool specification, and integration plan. Decided before any code is written, reviewed with your IT and security teams.

Build and test

MCP server deployment, custom tool development, integration with your AI platform, and testing against your own EPM scenarios with your team.

Training and handover

Hands-on training for users and admins. Documentation for the long term. Optional managed service for ongoing tool expansion.

Common questions

AI for Finance. Frequently asked.

What is MCP and how does it connect AI to Oracle EPM? +

MCP (Model Context Protocol) is an open standard that lets AI assistants like Claude and ChatGPT safely interact with external systems such as Oracle EPM Cloud. The FMEPM open-source MCP server acts as a bridge, exposing Oracle EPM capabilities (running business rules, reading substitution variables, exporting data slices) as tools the AI can call directly. This turns conversational requests into EPM actions, validated against your actual data.

Is the FMEPM MCP server for Oracle EPM open source? +

Yes. The public v1 is available at github.com/fmepm under the MIT license. It includes 7 tools covering core Oracle EPM Cloud operations: version discovery, listing applications, reading substitution variables, updating substitution variables, running business rules, exporting data slices, and job status polling.

Which AI assistants can I use with Oracle EPM through MCP? +

Any AI assistant that supports the Model Context Protocol can connect. Claude (Anthropic) has the most mature MCP support today. As MCP adoption grows, more assistants are adding support. The server itself is AI-agnostic. It exposes Oracle EPM operations as MCP tools, and any compliant MCP client can use them.

Is it safe to connect AI to Oracle EPM in a regulated environment? +

Safety depends on the deployment. The FMEPM MCP server respects Oracle EPM's native security model (no privilege escalation), authenticates through standard Oracle identity, and only exposes the specific operations you configure. For regulated finance environments, enterprise-grade AI platforms such as Amazon Bedrock, Azure AI Foundry, and Google Vertex AI offer Zero Data Retention (ZDR) and data residency controls. Part of the engagement is selecting the right deployment architecture for your compliance requirements.

What does an AI for Finance engagement typically include? +

A typical engagement covers: MCP server setup and configuration in your Oracle EPM environment, custom tool design for your specific workflows (month-end close, variance explanation, forecasting, ad-hoc reporting), integration with your chosen AI platform, security and governance setup, and team training for both finance users and admins. Ongoing support is available as part of a managed service tier.

How is this different from just using ChatGPT on exported EPM data? +

Copying EPM data into a public AI chat is brittle, manual, and often breaches data governance policies. MCP connects the AI directly to live EPM through a governed channel. The AI can pull data in real time, validated against your actual planning model, and take actions (like running a rule or exporting a slice) that the user has authorized. It is the difference between describing EPM to the AI and giving the AI the keys to operate EPM safely.