Give your finance agent a view beyond the ledger

A finance agent can tell you that electricity expense rose last month. The more useful question is what to investigate before the next bill: consumption, the contract, a regional price change, or a problem at a particular site.

That takes two kinds of evidence. Your accounting system holds invoices, contracts and budgets. deep.navy supplies outside context through one MCP connection: energy statistics, weather, company identity, economic indicators and news. The agent joins them into a briefing that a finance team can check.

Here is a workflow for a business with US facilities and overseas suppliers. The JSON examples are illustrative MCP tools/call parameters, not a report of live results. Your agent needs access to your approved internal files separately; deep.navy does not connect to bank accounts or accounting ledgers.

Start with a decision and a small input file

Ask for a weekly exception report: which costs or counterparties deserve a closer look, and what evidence would change the conclusion?

Give the agent a site list with coordinates, utility service class, monthly consumption, effective rate and contract renewal date. Add supplier legal names, countries and internal vendor IDs. A vendor ID keeps the agent from silently merging two businesses with similar names.

Keep each conclusion in one of three categories: an observed fact, a calculation from stated inputs, or a hypothesis awaiting confirmation. That distinction makes the report usable when somebody asks where a number came from.

Benchmark electricity expense against the right series

First, discover the supported EIA catalog:

{
	"name": "energy_search",
	"arguments": { "query": "electricity" }
}

Then request a bounded comparison window. This example selects California’s commercial sector, monthly average retail price:

{
	"name": "energy_fetch",
	"arguments": {
		"dataset": "electricity/retail-sales",
		"column": "price",
		"frequency": "monthly",
		"start": "2025-01",
		"end": "2025-12",
		"filters": [
			{ "facet": "stateid", "values": ["CA"] },
			{ "facet": "sectorid", "values": ["COM"] }
		],
		"limit": 12
	}
}

Choose dates that match your invoices, and verify the state and customer sector against the returned dimensions. Read the unit before calculating: cents per kilowatt-hour and dollars per kilowatt-hour differ by a factor of 100. Follow nextPageToken when present, preserving the other arguments. The energy guide explains missing values, source metadata and provider revisions.

EIA’s retail series provides a benchmark across customers, not the tariff in your contract. Its electricity data resources distinguish sales, revenues and average prices. A difference from that benchmark is a reason to inspect an invoice, not proof that a utility overcharged you.

For an initial invoice bridge, calculate the consumption effect as (new kWh − old kWh) × old effective rate, and the rate effect as new kWh × (new rate − old rate). Those two pieces reconcile the change in the energy charge. Track demand charges, taxes and fixed fees separately; folding them into a single rate can hide the actual cause.

Add weather to the next conversation

For each US site, call weather_forecast with its latitude and longitude and weather_alerts for active notices. These tools can help frame a facilities question: should we review cooling demand, staffing or a contingency plan?

A forecast alone does not quantify future consumption. Estimating that requires your site’s own temperature-to-load model and operating schedule. The NWS API documentation describes its point forecasts; our weather tools expose current forecasts and active alerts, not a historical forecast archive.

Use scenarios when that model is missing. With fictional inputs of 100,000 kWh and $0.15/kWh, a 10% consumption increase adds $1,500 before other charges. Label that an assumption-based sensitivity, not an EIA result or a weather prediction.

Give supplier reviews an identity check

Use company_search to find candidate legal entities, then company_fetch with the selected LEI. Confirm the match against your supplier record before using it. Search news with the confirmed legal name and fetch the original articles that matter.

GLEIF’s Level 2 data describes reported accounting consolidation relationships. It can help explain a group structure; it does not establish creditworthiness, sanctions status or complete beneficial ownership. A supplier without an LEI is an unresolved lookup, not an adverse finding.

For country context, discover an indicator with worldbank_search, then fetch the same years for each supplier country:

{
	"name": "worldbank_fetch",
	"arguments": {
		"indicatorId": "NY.GDP.MKTP.KD.ZG",
		"country": "MEX",
		"startYear": 2020,
		"endYear": 2024,
		"limit": 10
	}
}

This is annual GDP growth. It supplies background for a country review, not a prediction of an individual supplier’s cash flow. Keep observation years separate from retrieval dates, and leave missing values missing.

Ask for this briefing

After connecting your agent, provide the approved input files and paste:

Prepare a weekly finance exception brief from my site budgets, invoices and supplier list. Compare electricity invoices with matching EIA state/sector periods. Separate consumption, rate and other-charge changes. Add current NWS context for US sites without inventing a load model. Resolve supplier identities through GLEIF, flag ambiguous matches, and use news plus annual World Bank indicators for context. Return at most five issues, each with the internal record, external source URL, observation period, retrieval time, calculation or hypothesis, missing evidence, and the next question for the owner. Do not initiate payments or change budgets.

The deliverable is a short work queue: “check this demand charge,” “confirm this legal entity,” or “review this contingency assumption,” with enough evidence to act on the question. Preserve the source and untrusted-content metadata when passing it to another agent.

For investment research, continue with a filing-to-research workflow. For a market hypothesis built from weather and energy, try the energy research playbook.