)
You probably didn't get into finance to spend three days hunting down data across five different systems just to produce a travel expense report. Yet here we are.
Most finance teams are already using AI in some form. According to CFO Connect's 2025 Tools Report, 56% of finance leaders now use AI in their work. But there's a meaningful gap between using AI to draft an email and using AI to interrogate your actual spend data in real time. And there's a wider gap still between that and the future most coverage hints at: agents that surface a cash runway problem six months before it lands on your desk.
The Model Context Protocol (MCP) is what closes the first gap, and starts on the second.
This article explains what MCP actually is, how AI agents work in practice, and what it means for your spend management strategy. No technical background required.
Key takeaways:
MCP is an open standard — a single connector that gives AI workstations like Claude, Dust, and ChatGPT secure, real-time access to your finance tools, without custom integrations.
It works for every role on the finance team. Forecasts, AP chase-ups, bookkeeping, settlements, supplier intelligence, your AI assistant can run these inside the workstation your team already uses.
It replaces the export-and-paste workflow that's been the norm for AI in finance. Instead of pulling data into spreadsheets and uploading it, your AI tool reads the live data directly.
The platforms best positioned for MCP are the ones that have already consolidated spend data in one place. One source of real-time truth is what makes an AI assistant reliable.
What is the Model Context Protocol?
The finance-first definition
MCP is a connector that lets your AI assistant securely query other software in real time. Instead of logging into your spend management tool, exporting a spreadsheet, uploading it to ChatGPT and then asking a question, MCP connects your AI tool directly to the live data. You ask a question in plain language. You get a reliable answer in seconds, drawn from live numbers, not a static export.
)
Why the old way was broken
Before MCP, every AI-to-finance-tool connection was bespoke. Connecting your AI assistant to your accounting software took one piece of integration work. Connecting it to your spend platform took another. Each one was effort-heavy, brittle, and rarely in real-time unless you'd built and paid for a formal two-way integration.
The USB moment for AI
MCP replaces all of that with a single standardised protocol. The USB analogy is the simplest way to make sense of it: before USB, every peripheral device needed its own proprietary cable. USB standardised connectivity so any device could plug into any computer through one port. MCP does the same for AI — any MCP-compatible assistant can now access any MCP-ready finance system through a single connection, without a developer in the room.
Released by Anthropic in November 2024, MCP was adopted within months by OpenAI, Google, and Microsoft. The feature took off quickly inside the developer community, and in December 2025 Anthropic opened it up further, making it available as vendor-neutral infrastructure. For CFOs evaluating platforms, that means no lock-in risk: any tool that supports an MCP layer will plug into any AI workstation you choose.
Before MCP vs. after MCP
Before MCP: the finance data bottleneck. Finance teams have questions. Lots of them. But getting answers means exporting data from spend management tools into spreadsheets, manually filtering and analysing to find what you need, building reports for every new question, and waiting for someone else to build the report you need. The result is a reporting bottleneck: valuable insights delayed behind operational tasks, and AI tools that can't help because they don't have access to your actual finance data.
After MCP: your finance data, ready to talk. MCP connects spend management data to AI assistants like Claude, Dust, and ChatGPT, so finance teams can ask questions in plain language and get answers in seconds. Now your team can:
Ask: "What's our cash position?" or"Compare our annual travel spend year-on-year."
Get answers in seconds, drawn from real Spendesk data
Keep exploring with follow-up questions, no need to rebuild reports
Work in tools the team already uses — Claude, ChatGPT, or Dust
Overall, strategic insights stop waiting behind operational tasks. AI becomes useful because it's connected to real finance data. All while teams cut their reliance on exports and spreadsheets, answer quick and complex questions on the spot, and use AI safely without giving up control.
The building blocks of finance tech connectivity
)
Spend management tools like Spendesk connect your financial stack through three pathways:
Native integrations sync directly with accounting systems like NetSuite, Xero, and DATEV.
Open APIs connect to BI and reporting tools like Google Sheets.
The MCP connector gives AI tools — Claude, Dust, ChatGPT — live, read-only access to your spend data.
All three draw from the same source: every transaction your business processes, in one place.
The practical difference of having all three is immediate. A CFO asking "What's our committed spend versus forecast this quarter?" no longer needs an analyst to pull and reconcile the data. The answer comes back right away, followed by an option to update the forecast sheet and the board deck automatically.
Each layer does a specific job:
Integrations keep your books accurate.
API keeps your dashboards current.
MCP makes all of it accessible to the AI tools your team already uses, with no new builds required.
A note on data safety: an AI assistant can only access the data its user has access to. In practice, a team member with bookkeeping responsibilities in the spend management tool will only be able to query what they already see in the original system.
Examples of what finance teams can ask AI through an MCP connector
Role | What you can ask | What you get |
|---|---|---|
CFO / Finance Director |
| Strategic answers from live data, without waiting for a report to be built |
Controller |
| Real-time policy compliance visibility, without manual auditing |
FP&A Analyst |
| Trend analysis in seconds, without rebuilding reports from scratch |
Head of Accounting |
| Full payment-pipeline visibility without logging into multiple screens |
Procurement |
| Spend intelligence to support renegotiations and contract reviews, on demand |
MCP tools for finance: where the protocol meets your tech stack
What's already live
The MCP ecosystem has grown quickly. By early 2026, over 11,000 MCP servers were available in public directories. Stripe has published an official MCP server for payment operations. Digits launched a read-only accounting MCP server in April 2026, giving accounting practices direct access to client financial data inside AI tools. Microsoft Dynamics 365 Finance and Operations now supports MCP. The pace of adoption in spend management is a clear signal of where this is heading.
How Spendesk connects to your AI tools
For Spendesk users, MCP works like this: the Spendesk MCP server connects your spend data — payables, settlements, suppliers, purchase orders, and more — directly to your AI assistant.
You can connect through Claude Cowork (Anthropic's enterprise AI workspace), ChatGPT, or Dust — or any other AI platform built for company-wide workflows.
Each team member connects using their own Spendesk credentials, and sees data based on their existing Spendesk permissions. The Spendesk MCP is read-only by design — it can query, analyse, and surface information, but it can't create or approve payments, modify invoices, or change any Spendesk data. You stay in control.
Is your tech stack MCP-compatible? A quick checklist
When evaluating whether your existing tools can take part in an MCP ecosystem, four characteristics matter. You don't need to assess these yourself, but knowing what to ask your vendors is useful:
A structured REST API with clear data models. Systems with well-documented APIs connect easily; fragmented or undocumented APIs need significant work before MCP integration is viable.
OAuth 2.1 authentication. MCP requires this for secure agent access without exposing credentials. Legacy systems still using basic authentication will need upgrades.
Webhook support. Lets agents respond to real-time events (a new invoice received, an approval completed) rather than polling for updates — which dramatically improves response times.
Read-only access options. Granular permission scoping means you can deploy agents with minimal write access, preserving data integrity. The Spendesk MCP is built this way by default.
Most accounting software vendors haven't yet published official MCP servers, but the trajectory from early adoption to mainstream availability typically takes 18–24 months once major platforms commit.
Real-world use cases for AI agents in finance
Automated financial reporting with AI
Monthly reporting typically eats two to three days of capacity for a 150-person company: pulling data from multiple systems, reconciling discrepancies, compiling variance commentary.
With Spendesk MCP connected to your AI assistant, asking "What's our spend by department this month versus budget?" returns an answer in seconds, drawn from live data. Follow-up questions like "Which department is furthest over budget?" and "What's driving the variance in marketing?" can continue naturally — without rebuilding the query from scratch.
Use case 1: Real-time policy exception monitoring
The AI assistant watches transactions flowing through Spendesk and surfaces anything that falls outside policy — categories that shouldn't appear, approvals routed around the standard workflow, single charges over a defined threshold. The team gets a daily digest, or an instant flag for anything serious. The exceptions that used to be caught on a sample basis at month-end are caught in the moment, while the spend is still recoverable.
Use case 2: Weekly AP exposure briefing
Every Monday morning, the AI pulls together the week's AP picture: top exposures by supplier, payments stuck waiting for approval, anything overdue, anything large. It posts the briefing as a structured document the team can work from before the week starts. The cash conversation that used to begin with "let me pull that together" begins with "here's what we're looking at."
Use case 3. Quarterly supplier and SaaS audit
On demand, or on a quarterly schedule, the AI reviews vendor spend across the company and surfaces what's shifted since last quarter: top contracts ranked by leverage, suppliers that haven't been touched in six months, SaaS subscriptions that nobody's actually logging into. The kind of intelligence that usually only emerges when someone finds the time to go hunting for it, produced in the time it takes to make a coffee.
What this means for your spend management strategy
The consolidation question
The teams that will get the most from MCP-enabled AI are the ones that have already consolidated their spend data into a single platform. If invoices sit in one system, expense claims in another, and purchase orders in a third, an AI agent has no unified context to work with. It can only help with what it can see.
Three questions to assess your readiness
Is your spend data in one place? Smart company cards, invoice payments, expense claims, and purchase orders should flow through a single platform. If they do, AI agents can query a unified data set and give reliable answers. Learn more about how Spendesk connects your finance stack.
Are your approval workflows digital? Email and spreadsheet approvals can't integrate with AI agents. Digital approval workflows with clear routing rules create the structure agents need.
Does your platform integrate with your accounting software via API? Integrations with Xero, QuickBooks, Sage, or NetSuite mean spend data flows programmatically between systems — the foundation for AI that cross-references spend against financial records.
Start with what you already have
If you answered yes to all three, you're well positioned. Spendesk users who already run all company spend through the platform and connect to their accounting software are, in practical terms, already MCP-ready.
If you answered no to one or more, consolidation comes first. Build the data foundation, and the AI layer follows naturally.
Kevin Steele, a fractional CFO who built his own FP&A application in Claude Code, put it sharply on a recent CFO Connect session: "Anybody can make a fancy dashboard. What's really key is the financial engineering in the background that makes it actually work." That's the pattern: AI tools earn their place in a finance function not through interface, but through the data layer behind them. (CFO Connect, Building an AI-Native FP&A App in Claude Code)
Frequently asked questions
What is MCP and how does it work for finance?
The Model Context Protocol is an open standard that lets AI assistants connect securely to external tools and data sources. For controllers, FP&A analysts, and CFOs, it means your AI assistant (Claude Cowork, ChatGPT, or Dust) can access your live spend data and return reliable answers in plain language — without manual exports or custom integrations.
For Spendesk users: connect your account through your AI assistant's integrations settings and start querying your real payables, supplier, and PO data. The connection is read-only by design, so no data can be changed. Key capabilities include cash position, overdue AP, spend by department, payment status, invoice matching, budget variance, and supplier intelligence — all from a single conversational interface, connected to your real Spendesk data.
What finance tools support MCP integration?
Adoption is early but accelerating. Stripe, Digits, and Microsoft Dynamics 365 Finance have published official MCP servers. Most accounting software vendors haven't yet, but the trajectory from early adoption to mainstream availability typically takes 18–24 months.
How does agentic AI differ from traditional automation in finance?
Traditional automation follows fixed scripts that break when they meet exceptions. Agentic AI reasons over context — it can recognise that two differently named suppliers are the same vendor, or flag that a VAT rate doesn't match historical patterns for a given expense category.
The key difference isn't speed or scale. It's the ability to handle the grey areas that rule-based systems can't. Gartner predicts 33% of enterprise software will include agentic AI by 2028, up from less than 1% in 2024 — a measure of how rapidly this shift is moving. And critically, agentic AI still operates within constraints you define. Approvals, payments, and modifications stay under human control. The agent surfaces the information. You make the decision.
What spend management tools are compatible with AI agents and MCP?
Spend management platforms that consolidate all spend in one place — invoices, expense claims, smart company cards, and purchase orders — are best positioned for MCP integration. Spendesk's MCP server connects directly to Claude Cowork, ChatGPT, and Dust, giving your team conversational access to spend, payables, supplier, and PO data without exports or custom development. Spendesk also integrates with Xero, Sage, QuickBooks, and NetSuite, extending MCP-connected intelligence across your broader financial stack.
Curious how Spendesk works?
Try an interactive demo to see spend control and approvals end-to-end.
Get a free tour)
Maxime Reding
Originally from Luxembourg and educated at the Rotterdam School of Management, he brings a sharp analytical mindset and a background in strategic management to our finance team. Maxime joined Spendesk in 2021 and has grown steadily through the ranks since, becoming our resident finance engineer.
)
)
)
)
)