Why finance leaders should care about MCPs
If you are a Finance team evaluating planning software, this should be one of your first questions.
The attached screenshot shows a 100% accurate output from Claude. The data is sourced from a Finance data warehouse built in Microsoft Fabric and connected to Claude via the Power BI MCP interface.
The attached screenshot shows a 100% accurate output from Claude. The data is sourced from a Finance data warehouse built in Microsoft Fabric and connected to Claude via the Power BI MCP interface.
Fabric, Power BI, and Claude are all relatively familiar, but what role does the MCP play?
MCP (model context protocol) interfaces dictate how external data sources are connected to LLMs, which then make information available to end users.
This protocol is the data integration mode of the AI era and probably the biggest key to creating AI agents that are actually useful.
The MCP is a design standard, sort of like the REST standard for APIs. Each software provider can have their own MCP interfaces, similar to how they have their own APIs. Most LLMs will be set up to connect with any interface that follows the MCP standard.
The simplified flow looks something like this:
Source data + MCP interface + Business context + LLM = AI agent
Unlike APIs, MCP interfaces not only connect your data in real-time with LLMs (when then power your AI agent), they provide the LLM with your business context - data structure, operating model, design preferences, and more - so that the AI agent can provide consistently accurate and useful responses.
Providing context via the MCP is like providing instructions to a data engineer on how to pull the data and and an analyst how to analyze and present the data.
Providing this context requires business acumen and some technical aptitude, but it does not require AI infrastructure, capital investment, or a PhD in AI. It is not training in the same sense that LLMs are trained.
The MCP landscape is changing rapidly. Some MCP interfaces will be superior to others. Based on our experience, new interfaces are releasing, quickly improving, and leap-frogging each other, similar to the LLM race. MCP interfaces will vary in terms of how effectively they share data.
You can choose to buy pre-built AI agents with their own preset context, but we expect that many Finance teams will benefit from at least having the option to design and control the context that informs their agents.
Combining a solid data foundation, the right MCP infused with the right context, and the right LLM is the recipe for Finance teams to deliver insights in an AI-native fashion. If done correctly, a ton of ad hoc triaging and analysis can be wiped out, freeing up time for more strategic work.
These capabilities are arriving now, they are not years out. Now is the time for these concepts to start becoming more mainstream. Ryan Kim and our team at OVG are excited to continue sharing them with Finance organizations!
