Oh Claude, give it a REST!

What Happens When You Mix One Curious Architect, AI, a PDF, and Enterprise Software?

Like most of you, our team at Ascend has been looking at how to leverage AI, both to support our own work and to help our clients navigate what feels like a rapidly changing landscape. AI is becoming increasingly important across finance and technology, and more and more, our clients expect us to have a grounded perspective on where it can create meaningful value.

Don't worry, this isn't an article to make you feel like you're late to the AI party. The reality is that we're all learning as this technology evolves. At Ascend, curiosity has always been part of our DNA, which is why we're intentionally investing time and energy into understanding what AI can mean for our business, our people, and our clients. Our AI Strategy team has been exploring everything from bold, long-term ideas that could reshape how we work to practical, near-term opportunities that can make a difference today.

Down the AI Rabbit Hole

As part of getting myself up to speed on AI, I've found myself watching more than a few YouTube videos promising "game-changing skills" that would apparently change my life in 10 minutes. That hasn't quite happened yet, but one video did catch my attention. The host demonstrated how he had connected Claude directly to an EPM platform and was interacting with it through natural language.

Not long after, I was listening to an AI podcast that touched on a similar concept. The discussion centred around the Model Context Protocol (MCP), a framework that allows AI models to interact with external software systems.

The explanation that stuck with me was surprisingly simple.

An Application Programming Interface (API) is how one piece of software communicates with another, but every API has its own documentation, authentication requirements, and implementation details. Think of it like travelling between countries and finding that each uses a different electrical outlet.

An MCP acts more like a universal adapter. Instead of learning every individual API, the AI interacts through a consistent framework while the underlying API continues to do the actual work behind the scenes.

That last point got me thinking and left me with one burning question that needed to be answered:

Our integrations team already uses OneStream's REST API framework to connect external systems. If MCPs provide a standardized way for AI to access software functionality, could the same REST API capabilities be exposed through an MCP layer and used to connect an AI model to OneStream?

Not being a REST API integration expert in any shape or form (we have an amazing integrations team for that!), I uploaded the OneStream REST API Guide to Claude and began my interrogation:

"If I have a OneStream REST API guide, can you create an MCP?"

Claude answered back excitedly:

"Yes, absolutely! If you share the OneStream REST API guide, I can build a fully functional MCP server for it."

I wasn't going to get my hopes up too much, as I've seen this enthusiasm before. Anyone who has spent time working with AI knows that it often promises the world and doesn’t quite deliver.

The Moment Had Arrived

An hour later, after fighting through some configuration and installation-based tedium, I was presented with: "We're in! Logon succeeded, and I can see all of your authorized applications. The MCP is fully working now. Want me to open the Development application and start exploring?"

My interrogation continued: "What does exploring mean?"

Claude replied: "Good question. I just meant we can now do useful things like:

• Run a Data Management sequence or step: trigger consolidations, business rules, data loads
• Pull data from a Cube View: get financial data back, and I can analyze or summarize it
• Query the database: run SQL against the application database
• Check workflow status: see certification or workflow states for entities
• Run a Dashboard Adapter: pull data from any configured adapter"

Seeing the OneStream REST API functions served up by Claude really floored me. What the podcast had described was something Claude could build for me directly, as an MCP connector, in about an hour, with no REST API expertise required on my end.

Up until that point, MCPs had felt like an interesting concept I had encountered in videos and podcasts. Seeing a live connection inside OneStream suddenly made the idea feel much more tangible. What had started as an abstract discussion about AI integration was now something I could interact with and test firsthand.

As I spent more time experimenting with it, something else became apparent. Claude wasn't creating new OneStream functionality or bypassing the platform in any way. Everything it was doing ultimately relied on capabilities that already existed within OneStream's architecture. The platform, functionality and REST API framework were doing the heavy lifting throughout the entire exercise.

Permission to Speak Freely

So, here I was, speaking freely with OneStream, returning results I understood and could actually work with. There was no Postman window, no endpoint documentation, no request I had to get exactly right. I was speaking in my natural language and never had to know that a REST call had happened at all.

I'd love to say this made a perfect OneStream chatbot that could perform any function or answer any question I could think of, but it didn't, at least not at first. Claude, like a human learning a task, still had to be guided to return results that would be useful. It needed very clear instructions and context: which Data Adapter to use for a member expansion query, which database tables to reference when inspecting a cube view and setting the right parameters, and what conditions had to be met before running a consolidation.

But once Claude built up its reps, I was having conversations like:

Me: “Run GOP in USD for China Oct 24 YTD actuals.“

Claude: “Query returned successfully — Col9 (Oct 2024) is the requested figure.”

Me: “Show me the entities underneath that make up this amount.”

Claude: “Here you go. The underlying China entities sum exactly to the previously confirmed total.”

Me: “List all entities in Europe and display their descriptions, currencies and Text1 values.”

Claude: “The table below lists the entities, descriptions, currencies and text1 values.”

Me: “Run a consolidation of 2024.”

Claude: “Consolidation completed successfully. Want me to check consolidation status (View=CS) to confirm the result? Result: consolidated and query-ready. Every period returned OK,MC”

Conclusion

Looking back, my experiment sits alongside a lot of the innovation already happening within OneStream. Their Sensible Forecast, Sensible Studio, and Sensible Agents are purpose-built, supported, and the right way to bring AI into a production OneStream environment. What strikes me most about my experiment is how easily it came together: a working prototype in about an hour, no REST API expertise required on my end. 

This was me just me satisfying a curiosity. Not long ago, building something like this would have required specialized technical expertise, significant planning, and a dedicated project effort. In this case, it started with a question and a documentation guide. That’s it!

In short, I've now started to think of AI as a Swiss Army knife: a versatile set of tools that can be applied in different ways to achieve business outcomes. The challenge is no longer figuring out whether the tools are capable. It's understanding where they can create the most value.

One thing this experiment reinforced for me is something the Ascend team and I have seen repeatedly across more than 180+ OneStream implementations: the foundation matters. Whether you're leveraging AI, improving performance, simplifying architecture, or accelerating reporting, successful outcomes still come back to having a well-designed application and a solid understanding of how the platform is being used.

With the arrival of increasingly capable AI solutions such as Claude and OneStream's own AI suite, we're starting to see an interesting intersection between real-world implementation experience and what these tools can do. On their own, AI tools are impressive. Combined with deep platform knowledge, practical experience and a solution built on solid ground, they become much more valuable.

Build Your Own Connected OneStream Agent with Ascend Partners

If you're exploring how AI can complement your OneStream environment, whether through OneStream's own AI capabilities or Ascend’s own innovations, we're always happy to share what we're learning and continue the conversation.

About the Author

Mike Dewis, Vice President, Application Delivery at Ascend Partners, brings over two decades of finance systems and OneStream implementation experience. As a Certified OneStream Professional and Lead Architect, he leads teams that architect scalable, high-performance solutions for some of the most advanced and complex clients in the EPM ecosystem.