Dynamics Community Forum Thread Details

Using MCP for Large Sales Data Queries

I’m using MCP to analyze sales data (e.g., total sales for 2025, Q1 sales, and comparisons with current data).

To handle large datasets, I added instructions in the Overview section to:

Fetch records in batches (e.g., 1500 at a time)
Continue automatically for all batches
Aggregate totals across batches

(this is from MCP Overview page)

Problem:

First batch works correctly
Copilot asks to continue instead of proceeding automatically
Next batch often fails with “too much data”
Results are incomplete and inconsistant

Question:

What is the recommended approach for large datasets?
Does MCP support pagination/aggregation via Overview instructions?
How can we ensure automatic batch processing without prompts?
Should this be handled outside MCP (e.g., APIs or pre-aggregated data), and if so, can we process/aggregate the data externally and then load or expose the summarized data back into MCP for querying? If yes, how to do it?

Categories:

Dynamics 365 Business Central

Hello,
don’t rely on MCP Overview instructions to do reliable pagination/aggregation over large Business Central datasets; those instructions can guide the agent, but they don’t guarantee a loop will continue silently, avoid token/context limits, or aggregate every page consistently. The better pattern is to expose a purpose-built tool/API that returns the answer already aggregated, for example get_sales_total(year), get_sales_by_period(startDate,endDate), or get_sales_comparison(...), and let that backend handle Business Central API pagination using @odata.nextLink, batching, retries, and totals; then MCP only receives the small summarized result. For BC specifically, I’d either create an API page/query/codeunit endpoint that returns pre-aggregated sales totals, or use an external service/Azure Function/Power Automate/custom connector that reads BC in pages, aggregates externally, stores results in Dataverse/SQL/Business performance analytics, and exposes those summaries back to Copilot/MCP as a small tool result. Microsoft’s guidance for MCP timeout/large-data issues is also to focus queries on aggregated data, limit time ranges, and break complex analysis into steps, so raw transaction-level analysis through chat is the wrong layer for this.
Regards,
Oussama Sabbouh

Quick Links

Leaderboard > Small and medium business | Business Central, NAV, RMS