Your next finance app might not be an app...
- 18 hours ago
- 6 min read
For years, improving your finance stack mostly meant adding more tools. A software for accounting, another for expenses, another for cash management, another for collections... And often, a spreadsheet to connect everything in between.
The arrival of AI agents could start to change this logic. With platforms like Dust and the rise of the Model Context Protocol (MCP), the challenge is no longer simply which new application to add. It now becomes possible to directly query in natural language the tools already used by the company.
That shift is particularly relevant for companies that already have a solid finance stack.
Dust: a layer on top of your existing tools
Dust is not an accounting software, nor a cash or reporting tool. It is a platform that allows you to create AI agents directly connected to the company's internal data and tools.
Dust can work with different AI models and connect to external business applications via MCPs. And, for a finance team, the difference can quickly become significant.
Today, to understand why a department's expenses have increased, you generally have to open a tool, select a period, filter the data, possibly export a file and then analyze the result.
Tomorrow (and in some cases, already today) the question may simply become: "Why have current expenses increased over the last three months?" The agent will then retrieve the data they are authorized to access and provide the analysis directly.
MCP: three letters to remember
Behind this evolution lies, in particular, the Model Context Protocol , or MCP. This is an open standard initially developed by Anthropic that allows an AI assistant to connect to external systems and access, with the necessary permissions, the data and functionalities they make available to it.
Put simply, MCP creates a bridge between AI and your business applications. And the topic is already becoming essential in finance as well.
Pennylane now offers its own MCP server and specifically lists Dust among its compatible AI assistants. Users can grant their agents direct access to certain accounting and financial data stored in Pennylane, such as invoices, transactions, and VAT returns, depending on their assigned permissions. Access is currently read-only, although Pennylane is working on write capabilities as well.
For a business owner, this opens up very concrete uses. Instead of starting by searching for information in their accounting software, they can ask their agent using the data actually present in Pennylane.
Spendesk has also launched its MCP (Management Control Platform) to allow users to query expense data from assistants like Claude or Dust. This allows a finance team to search for invoices, compare expenses by cost center, or analyze suppliers without first exporting the data to Excel.
Stripe is moving in the same direction. Its MCP server allows an authorized AI agent to directly access certain Stripe data: balances, invoices, payments, customers, subscriptions, and even disputes. For a SaaS or eCommerce company the benefit is easy to understand: some business and financial information can be queried directly from the agent, without having to navigate the Stripe dashboard first.
Upflow is following the same path for accounts receivable. Its MCP server can be used to query invoices, payment history, and customer payment delays from an AI assistant.
Expenses, payments, subscriptions, customer receivables: gradually, several building blocks of the finance stack are becoming directly accessible to agents.
READ ALSO
What really changes for the leader
The point is not to chat with your accounting data just because you can. The real value is reducing the number of steps between a business question and an actionable answer.
For example, a business owner could ask:
"Which customers represent the greatest risk to cash collection today?"
"Which SaaS expenses have increased the most since January?"
"Which subscriptions saw the most growth this quarter?"
"Which suppliers account for the largest share of our marketing expenses?"
"What is the amount of the payments currently in dispute?"
This information already exists somewhere. However, it's often spread across multiple tools, dashboards, and sometimes even teams. The agent then becomes an access layer to the existing stack. And that's probably where the real change lies.
Should you therefore stop adding apps?
No, obviously not... An AI agent can never replace the quality of the tools it connects to. Therefore, if invoices are incorrectly filled out, expense categories are inconsistent, customer data is incomplete, or financial flows are poorly integrated, the AI will not magically produce reliable answers.
Accounting remains the source of truth of accounting, and therefore the most reliable source for your figures. Expense management software continues to control spending. Stripe continues to manage payments and subscriptions. The collections tool continues to organize the tracking of receivables.
AI sits on top of these building blocks to make information easier to access, combine certain data points, and speed up analysis.
A weak finance stack connected to an excellent AI agent is still… a weak finance stack.
READ ALSO
And what about the security of financial data?
This is obviously not something businesses should treat lightly. Connecting an agent to the company's finance system involves precisely defining who can access what, what data can be output from each application, and what actions can be performed.
Pennylane, for example, has implemented several safeguards for its MCP. Authentication is handled via OAuth, and the AI assistant inherits the exact same rights as the user who logs it in. If that user doesn't have access to a folder or data in Pennylane , their assistant won't have access to it either.
Access can also be revoked at any time, and currently, the Pennylane MCP operates in read-only mode, as mentioned above. The assistant can view authorized data, but cannot modify, create, or delete it. Furthermore, the data stored by Pennylane remains hosted in Europe, and Pennylane logs access within its own perimeter.
But one distinction is essential: once data is transmitted to an external AI assistant, the way that data is handled depends on that provider, not Pennylane.
Businesses therefore need to look at where the AI provider hosts data, how long it retains it, whether it can reuse the data, its GDPR safeguards, and its broader security controls. Depending on the AI assistant selected, some data may also be transferred outside the European Union.
Stripe also uses OAuth to connect its MCP clients, while Upflow restricts access through permissions Dust, for its part, offers governance and access-control capabilities.
In other words, connecting an AI to your finance stack should never mean giving it unrestricted access to everything. The security of the finance stack now depends both on the applications that hold the data and on the AI assistant the business chooses to share that data with.
The best finance stack might be the one you open less
For a long time, financial software was mainly evaluated through its interface: dashboard, graphs, filters, ergonomics... With AI agents, this logic should gradually change.
If a business owner can ask, "What are the three financial topics I need to look at this week?" and by obtaining a response constructed from its expenses, receipts, subscriptions or customer receivables, it will spend less time searching for information.
But saving time is only the first step. The real added value comes when this information makes it possible to identify a margin drift, a customer who is taking longer to pay, an unusual increase in costs or an impending cash flow problem.
In other words, a good finance stack shouldn't just produce more data or dashboards. It should enable the right signals to surface at the right time, so that you can act before the problem becomes unmanageable several weeks later.
Applications are not going to disappear. They will continue to collect, structure, and secure data. But their interface could become less central.
Before adding a new app to your finance stack, the right question may soon be:
Am I really missing a tool… or a better way to turn my existing data into better decisions?
The Blendy perspective
AI can already flag declining margins, late payments, or unusual increases in spending. It can even suggest an initial response: raise prices, cut certain costs, postpone a hire, or accelerate collections.
But financial decisions rarely have only one consequence.
Raising prices may improve margins, but it can also affect VAT collected, profitability, cash flow, and ultimately taxable income. Postponing a hire may protect cash in the short term, but it could slow down a project or limit operational capacity. Increasing investment may support growth while temporarily reducing available cash.
This is where accounting advisory becomes essential.
At Blendy, international CPA, based in Paris, Montréal and Miami, our role is to help business owners look at a decision as a whole: its impact on profitability, cash flow, tax, investment capacity, hiring plans, and the company’s ability to fund its growth.
Managing a company's finances isn't just about "What do the numbers say?". The important thing is to be able to take the best decisions to meet the company's objectives: "If I make this decision, what impact will it have elsewhere in my company?"
AI can surface the right information faster. Blendy provides the perspective, guidance, and overall vision needed to understand the broder financial consequences of a decision before making it.
Sources:
With Blendy , International chartered accountant: take advantage of all the benefits of digital accounting and international financial advice to accelerate your finance process and grow your business.
Certified by Pennylane , Dext , QuickBooks and Stripe , we support digital companies, eCommerce, IT services companies, SaaS companies, in France and internationally.
