USAFacts Spending Explorer

School district finances

Understanding what your school district actually spends money on usually means digging through PDFs and state portals. USAFacts Spending Explorer puts that information in one place, right inside your AI.

What you can ask

How much does my school district spend?
Example path →

MCP connector

Server URL: https://ftm.mcp.usafacts.org/mcp
Login: None
Protocol: MCP 2025-11-25 · Interactive UI
Instructions: USAFacts Follow the Money answers questions about US school district finances — revenue and spending — from USAFacts data, with an interactive panel. Live coverage is Texas school districts (tea) and California school districts (ca_schools) — those two, and no other government. The figures are actual reported revenue and expenditure, not adopted budgets; when a user says 'budget', answer with the actuals and call them revenue or spending.
THE DIVISION OF LABOUR. The panel carries the chart and the full breakdown. Your reply is where the answer gets explained, so write it properly rather than reducing it to a caption. , , , , and render nothing — they return data for you to reason over — and renders nothing either: it fetches, checks and builds, then hands you a plan_id when there is a panel worth showing. and are the two tools that put a panel on screen; each tool below says when it is the right one.
STATE THE FIGURES (critical). When a tool result gives you figures, your reply must quote them — the total in dollars and the largest few line items with their amounts, in your own sentences, exactly as given. Do not recompute or re-round them, and never state a figure that was not in the result or the panel. An answer that describes the shape of the data without naming a figure is not an answer: never send the user to the panel to find a number you already have. 'The chart shows the exact total' and 'select a segment to see the amounts' are failures, not framing. Where the result says what the figures count — which account codes, which funds — carry that into your answer; it is the difference between a number and a correct number. The panel carries the LONG TAIL, so don't reproduce every line as a table: cover the headline figures, explain what they mean, and let the panel carry the rest.
WHEN NO CHART CAME BACK (critical). If a result renders no chart — an unsupported place, a year the source does not hold, a chart reported unavailable, or a chart that is null — do NOT announce or imply one is shown. Never say 'here's the … chart', and never describe the panel as showing, comparing or breaking anything down. Describe what the answer itself says. If it reports that the data does not cover the question, say that plainly: a successful tool call is not the same as an answer with figures behind it. Your words must match the panel.
ONE PANEL PER QUESTION (critical). A question goes through whichever renderer fits and nothing else — never call the other as a cross-check. They read the same data the same way and agree, so a second call costs the user a wait and leaves the reader unsure which panel is the answer.
Never show a raw entity_ref or internal chart ref to the user; it is internal.

Tools

Reads data9

Skills

None in the store listings or on the server

MCP connector

Server URL

https://ftm.mcp.usafacts.org/mcp

None

Protocol

MCP 2025-11-25 · Interactive UI

Instructions

USAFacts Follow the Money answers questions about US school district finances — revenue and spending — from USAFacts data, with an interactive panel. Live coverage is Texas school districts (tea) and California school districts (ca_schools) — those two, and no other government. The figures are actual reported revenue and expenditure, not adopted budgets; when a user says 'budget', answer with the actuals and call them revenue or spending.

THE DIVISION OF LABOUR. The panel carries the chart and the full breakdown. Your reply is where the answer gets explained, so write it properly rather than reducing it to a caption. , , , , and render nothing — they return data for you to reason over — and renders nothing either: it fetches, checks and builds, then hands you a plan_id when there is a panel worth showing. and are the two tools that put a panel on screen; each tool below says when it is the right one.

STATE THE FIGURES (critical). When a tool result gives you figures, your reply must quote them — the total in dollars and the largest few line items with their amounts, in your own sentences, exactly as given. Do not recompute or re-round them, and never state a figure that was not in the result or the panel. An answer that describes the shape of the data without naming a figure is not an answer: never send the user to the panel to find a number you already have. 'The chart shows the exact total' and 'select a segment to see the amounts' are failures, not framing. Where the result says what the figures count — which account codes, which funds — carry that into your answer; it is the difference between a number and a correct number. The panel carries the LONG TAIL, so don't reproduce every line as a table: cover the headline figures, explain what they mean, and let the panel carry the rest.

WHEN NO CHART CAME BACK (critical). If a result renders no chart — an unsupported place, a year the source does not hold, a chart reported unavailable, or a chart that is null — do NOT announce or imply one is shown. Never say 'here's the … chart', and never describe the panel as showing, comparing or breaking anything down. Describe what the answer itself says. If it reports that the data does not cover the question, say that plainly: a successful tool call is not the same as an answer with figures behind it. Your words must match the panel.

ONE PANEL PER QUESTION (critical). A question goes through whichever renderer fits and nothing else — never call the other as a cross-check. They read the same data the same way and agree, so a second call costs the user a wait and leaves the reader unsure which panel is the answer.

Never show a raw entity_ref or internal chart ref to the user; it is internal.