Reading this with software, or want an AI to analyse it? lunenburgbudgetproject.org/ask has the prompt, the data API and an MCP server.

Updated September 5, 2026— the archive moves off the site’s host

This is how you query the data

Copy this. Paste it into Claude, ChatGPT or any assistant that can fetch a web page. Type your question at the end.

or select the text below — 3,212 characters

I want to look into the Lunenburg, Massachusetts town and school budget using an
independent public archive. Please fetch these and use them to answer my questions.

START HERE — what exists, and how to ask it something:
  https://lunenburgbudgetproject.org/llms.txt
      start here — what this site holds and how to read it
  https://lunenburgbudgetproject.org/api/index
      every machine-readable address, as JSON
  https://lunenburgbudgetproject.org/api/tables
      what data EXISTS before you ask anything: 49 datasets, the years each covers, row counts. 5KB
  https://lunenburgbudgetproject.org/api/schema
      the grain of every table, and four ways to get a confident wrong answer out of it
  https://lunenburgbudgetproject.org/api/questions
      107 worked questions, each with the query that answers it
  https://lunenburgbudgetproject.org/api/query?sql=SELECT%20name%20FROM%20sqlite_master%20WHERE%20type%3D%27table%27
      one read-only SELECT over the database, as a plain GET. Returns the documents its rows came from

FILES THAT ANSWER A PARTICULAR QUESTION:
  https://lunenburgbudgetproject.org/minutes/INDEX.txt
      what any town board discussed, and when — zoning, conservation, health, schools, planning, cemeteries, the library, anything
  https://lunenburgbudgetproject.org/data/rate-register.csv
      what does the town charge, and when did it change
  https://lunenburgbudgetproject.org/data/athletic-fee-schedule.csv
      athletic fees, by year and tier
  https://lunenburgbudgetproject.org/docs/analyses/free-cash.md
      is the town hoarding free cash, or rebuilding it — and what it would buy
  https://lunenburgbudgetproject.org/data/free-cash-proof.csv
      free cash — how much the town can spend without raising taxes, and nine towns to compare it with
  https://lunenburgbudgetproject.org/data/budget-lines.csv
      any budget line, any year or scenario
  https://lunenburgbudgetproject.org/docs/analyses/show-your-work.md
      HOW a figure was calculated — the formula, the inputs, and what is assumed
  https://lunenburgbudgetproject.org/data/model.json
      every figure this site computes, and how
  https://lunenburgbudgetproject.org/data/sources.json
      every source document, with a hash
  https://lunenburgbudgetproject.org/docs/analyses/athletics-ledger.md
      athletics: what it costs and who pays
  https://lunenburgbudgetproject.org/docs/analyses/sped-and-the-curve.md
      special education as a cost driver

TWO THINGS TO GET RIGHT, because the data lets you get them wrong quietly:

  1. The archive holds both what the town BUDGETED and what it SPENT. They differ by up to 59% on some lines, and a rate measured from one to the other is partly growth and partly the gap between them. Never mix them in one calculation. A budget line is also NET of grants, fees and state aid — it is not what a thing costs.

  2. Cite the document. Every figure here traces to a published document with a
     sha256, and the query endpoint returns the documents its rows came from. If an
     answer cannot be traced to one, say so rather than giving me the number.

If a query fails, that is a rate or cost limit, not missing data — say which.

My question:

It carries every address the assistant needs — the whole database, the budget lines, the town ledger, the annual town reports, and 1,422 agendas and sets of minutes across 40 town boards, 2025-01-06 to 2026-11-17 — so it reads the data instead of guessing at it, and cites the document each figure came from. You do not need to know any SQL, and you do not need an account anywhere.

Why the prompt has to include the links

Assistants generally refuse to fetch a URL they have not been given — a sensible rule, since it stops a web page from talking an assistant into visiting somewhere else. The side effect is that an assistant can read this site's front page, correctly describe everything it offers, and still not be allowed to open any of it. A link you paste yourself is authorised by you. That is the whole trick, and it works today with every assistant.

The addresses, as links

The same ones as in the prompt. Follow them directly if you would rather.

/llms.txtstart here — what this site holds and how to read it
/api/indexevery machine-readable address, as JSON
/api/tableswhat data EXISTS before you ask anything: 49 datasets, the years each covers, row counts. 5KB
/api/schemathe grain of every table, and four ways to get a confident wrong answer out of it
/api/questions107 worked questions, each with the query that answers it
/api/query?sql=SELECT%20name%20FROM%20sqlite_master%20WHERE%20type%3D%27table%27one read-only SELECT over the database, as a plain GET. Returns the documents its rows came from
/minutes/INDEX.txtwhat any town board discussed, and when — zoning, conservation, health, schools, planning, cemeteries, the library, anything
/data/rate-register.csvwhat does the town charge, and when did it change
/data/athletic-fee-schedule.csvathletic fees, by year and tier
/docs/analyses/free-cash.mdis the town hoarding free cash, or rebuilding it — and what it would buy
/data/free-cash-proof.csvfree cash — how much the town can spend without raising taxes, and nine towns to compare it with
/data/budget-lines.csvany budget line, any year or scenario
/docs/analyses/show-your-work.mdHOW a figure was calculated — the formula, the inputs, and what is assumed
/data/model.jsonevery figure this site computes, and how
/data/sources.jsonevery source document, with a hash
/docs/analyses/athletics-ledger.mdathletics: what it costs and who pays
/docs/analyses/sped-and-the-curve.mdspecial education as a cost driver

What to ask it

There are 107 questions this archive can answer, in plain English, each one already run against the data — so nothing on the list is a question we merely hope is answerable. Many carry a second line saying what the answer does not tell you, which is usually the more important half.

What you can ask this archive (PDF) — the list, in plain English, no queries
The same list with the query behind each one

If your assistant supports MCP

There is a server at lunenburgbudgetproject.org/mcp — add it as a remote MCP server in your client's settings, not in the chat. It gives the assistant eight tools over the same data, each one shaped so the mistakes this archive documents cannot be made: budget history takes one stage at a time, so a growth rate cannot be measured from an actual to a budget.

Every address in the prompt is generated from the same list that produces llms.txt and the index of published addresses, so the three cannot drift apart.