NetSuite MCP Agents

The problem

Month-end close is still, for most finance teams, a manual archaeology project. An accountant pulls a P&L and Balance Sheet for the current and prior period, eyeballs every line for anything that moved, and opens a half-dozen tabs to trace the ones that look off. Multiply that across subsidiaries and a controller can lose days of the close cycle to work that is repetitive and hard to onboard new staff into. The same pattern shows up twice more outside flux review: revenue that quietly slips through the cracks of a subscription book, and liabilities that exist in the business but haven't made it into the general ledger yet.

The approach

NetSuite MCP Agents bundles three agent-powered financial-analysis skills into a single client-side toolkit. It consumes NetSuite's native MCP APIs — read-only, never writing back to the ledger — and orchestrates Claude to reason over the data the way a senior accountant would. The same skill set deploys as a Claude Code Agent Skill, a Claude Project, or a Custom GPT, so it fits wherever a finance team already works.

Underneath, the toolkit combines SuiteQL for targeted retrieval, custom SuiteScript tools built on N/API 2.1, and a deterministic Python + openpyxl export step — Claude drives the analysis and narrative, but the final workbook is generated in code, not by the model, so the numbers are always reproducible.

The three agents, hands on

Three skills share that architecture. Each demo below is a faithful mock of the real agent's flow — every customer, vendor, subsidiary, account, and dollar figure is invented for illustration.

Revenue leakage

The leakage agent walks a subscription book schedule by schedule, reconciling what the contract says should be billed and recognized against what billing actually generated — contractual escalations that never reached the rate card, metered usage that never became an invoice, recognition schedules paused and forgotten. Run the scan, then click a flagged row for the agent's finding and its recommended fix:

Unrecorded liabilities

The liabilities agent hunts for expenses the business has already incurred that haven't hit the ledger yet. It cross-references three sources — vendor bills received but not approved, open purchase orders with item receipts and no bill, and each vendor's historical billing cadence — and assembles the candidates into a close-ready accrual entry. Untick any candidate and the proposed entry recomputes:

Variance (flux) analysis

The flux agent runs a seven-phase flow: discover the subsidiaries and books, pull the comparative P&L and Balance Sheet, flag moves above materiality (P&L ≥$50K and ≥10%; Balance Sheet ≥$100K and ≥10%), drill into the flagged accounts with targeted SuiteQL, classify each variance as Timing / Permanent / Classification / Unknown with a confidence score, export a ten-tab workbook deterministically, and surface the summary. The demo replays that run as the agent's console transcript:

The impact

Across pilot portfolio companies, the flux, leakage, and liabilities skills have compressed the analytical portion of month-end close from a multi-day effort to a same-day review pass — the agent produces the first read of every account, so staff spend their time validating narratives rather than writing them from scratch. For smaller finance teams, that has meaningfully deferred the need to add close-cycle headcount even as transaction volume grows, and because every export regenerates deterministically from the same source data, it has held up to audit scrutiny as well as a manually built workbook.