Microsoft just gave Dynamics 365 Finance and Operations something finance teams have wanted for years: the ability to ask a question about your ERP data in plain English and get an answer back, no report builder, no DAX, no ticket to the BI team. It is called the Dynamics 365 ERP Analytics MCP server, and it turns Business Performance Analytics into a conversational AI analyst you can talk to from Copilot Studio or VS Code.
For any CFO or controller who has waited three days for a variance breakdown, that sounds close to magic. It is a genuine step forward. It also has hard edges that matter the moment you move from exploring numbers to closing the books on them. This guide covers what the ERP Analytics MCP actually does, the limits Microsoft documents, and the reporting work it does not replace.
What the Dynamics 365 ERP Analytics MCP actually is
The Model Context Protocol (MCP) is an open standard for connecting AI agents to data systems. Microsoft’s ERP Analytics MCP server is an implementation of that standard that sits on top of Business Performance Analytics (BPA), the analytics layer bundled inside Dynamics 365 Finance and Operations.
When you ask a question, the agent reads the BPA data model, writes the DAX query for you, runs it, and returns the result as structured data an assistant can explain. You never see the query. You describe the insight you want, and the agent handles the mechanics. It works in Microsoft Copilot Studio and in Visual Studio Code with GitHub Copilot, and it can run alongside the separate Dynamics 365 ERP MCP server that performs actions inside F&O.
The data it can reach spans the three value chains BPA models:
- Record-to-Report: financial data, P&L, budgets.
- Procure-to-Pay: purchase orders, vendor management.
- Order-to-Cash: sales orders, invoicing, receivables.
What it does well
For ad-hoc, exploratory questions, this is a real capability, and it is worth being fair about how useful it is. The kinds of questions it handles well, straight from Microsoft’s documentation:
- Aggregations: total sales by region last quarter.
- Trend analysis: revenue over the past 12 months.
- Comparisons: vendor performance Q1 versus Q2.
- Rankings: top 10 customers by revenue.
- Anomaly spotting: invoices that deviate from normal patterns.
It respects row-level security, so a user only sees data their F&O and BPA roles allow. For a controller who wants a quick read before a meeting, or an FP&A analyst pressure-testing a hunch, it removes a real bottleneck. This is the part of the announcement that deserves the attention it is getting.
What it can’t do (the part the demos skip)
The ERP Analytics MCP is an exploration layer, not a reporting or consolidation layer. Six documented constraints define where it stops.
1. It refreshes twice a day, not in real time
BPA data currently refreshes on a 12-hour cycle, and the full pipeline of sync, transform, and refresh can take four to five hours or more depending on volume. Hourly refresh is listed as in development. For strategic trend questions that is fine. For anyone reconciling intraday, watching a warehouse during month-end, or acting on today’s cash position, twice-daily data is a different tool for a different job.
2. It answers questions, it does not consolidate
A conversational answer about revenue is one thing. A consolidated, audit-ready group statement is another. Multi-entity financial consolidation with intercompany eliminations, currency translation, and minority interest is a modeling and controls job, not a natural-language query. The MCP surfaces what BPA already holds. It does not perform the elimination logic that turns six legal entities into one board pack.
3. A chat answer is not a governed report
Because the agent generates DAX on demand per question, two people phrasing the same question slightly differently can get answers that are hard to reconcile line for line. That flexibility is the point for exploration. It is also why a conversational result is not the same as a repeatable, version-controlled statutory report where every figure traces to a journal entry the same way every close. Governed reporting needs a fixed, auditable definition. Conversation needs the opposite.
4. It only sees what BPA models
The tables, measures, and dimensions available depend entirely on your Business Performance Analytics configuration, scoped to those three value chains. Custom financial dimensions, operational metrics outside R2R, P2P, and O2C, and anything you have not modeled into BPA are simply not reachable. If your reporting depends on data BPA does not carry, the agent cannot answer for it.
5. Query limits rule out line-level extracts
Each query has a 120-second timeout and returns roughly 10 MB of data. Microsoft’s own guidance is to use aggregations and summaries rather than requesting detailed transaction-level data. That is the right design for a chat interface. It also means the MCP is not the way you pull a full transaction ledger for a deep audit or a drill-down that ends on a specific voucher line.
6. It is public preview, with prerequisites
The ERP Analytics MCP server is in public preview, which Microsoft notes is not intended for production use and may have restricted functionality. It requires Dynamics 365 Finance and Operations version 10.0.38 or higher and Business Performance Analytics version 2.4.3224 or higher, both installed and fully refreshed. Plenty of finance teams are not on those versions yet.
The real question for finance leaders
The useful way to frame this is not “AI analyst versus reporting platform.” It is exploration versus governed reporting, and every finance function needs both.
| The job | ERP Analytics MCP | A purpose-built reporting layer |
|---|---|---|
| Ad-hoc, conversational questions | Strong | Not the point |
| Governed statutory statements | No | Yes |
| Multi-entity consolidation and eliminations | No | Yes |
| Near real-time operational views | Twice daily | 15 to 30 minutes |
| Drill from group figure to journal line | Limited by query size | Yes |
| Data scope | What BPA models | Your full F&O data model |
The two are complementary. In fact, a clean, well-modeled reporting foundation makes conversational agents better, because they are only ever as good as the data underneath them. The teams getting the most from AI in finance are not choosing between a chatbot and a reporting platform. They are pairing exploration on top with governed, real-time reporting underneath.
Where Metrixs fits
Metrixs is the governed reporting layer purpose-built for Dynamics 365 Finance and Operations. It reads your F&O data through Azure Synapse Link and rebuilds it into clean, reporting-shaped models, then ships the parts the MCP leaves out:
- Pre-built consolidation with intercompany eliminations and multi-currency translation across every legal entity.
- Refresh every 15 to 30 minutes, not twice a day, for teams that reconcile and operate intraday.
- Drill-down from any group figure to the originating journal line, with a governed definition that reads the same every close.
- Live in under six weeks, with 1,000+ metrics and 100+ pre-built reports, and 99.9% accuracy against the ledger.
Ask the MCP what happened last quarter. Rely on a governed layer to close, consolidate, and report the numbers your board and auditors sign off on. If you want to see what that foundation looks like on your own D365 data, book a Metrixs demo, and compare it side by side with what native analytics ships today.
Related reading: Do you still need a third-party analytics layer if you have Microsoft BPA? and Financial close automation for D365 F&O.
Frequently asked questions
Is the Dynamics 365 ERP Analytics MCP generally available?
No. As of early 2026 it is in public preview, subject to Microsoft’s preview terms and not intended for production use. The separate Dynamics 365 ERP MCP server that performs actions reached general availability, but the analytics server is still preview.
Does the ERP Analytics MCP replace Business Performance Analytics reports?
No. BPA reports give fixed, pre-built dashboards. The MCP adds conversational, on-demand questions on the same data. They are two ways to read one dataset, and neither performs multi-entity consolidation.
Can it consolidate multiple legal entities?
No. It answers analytical questions on what BPA already holds. Consolidation with intercompany eliminations, currency translation, and a defined group structure is a separate reporting job that a purpose-built layer handles.
How fresh is the data?
It refreshes twice daily on a 12-hour cycle, with a pipeline that can take four to five hours. Hourly refresh is in development. It is not a real-time or intraday tool today.
What do we need to run it?
Dynamics 365 Finance and Operations 10.0.38 or higher, Business Performance Analytics 2.4.3224 or higher, appropriate BPA and F&O security roles, and a connection from Copilot Studio or VS Code. Row-level security from BPA is enforced throughout.