Nber Macrohistory MCP server

@pipeworx/nber-macrohistory

NBER Macrohistory Database — the National Bureau of Economic Research's public archive of long-run U.S. (and some international) macroeconomic series, most running from the 1800s through the 1960s, across 16 subject categories (prices, money & banking, production, construction, employment, agriculture, foreign trade, government finance, and more). Look up a specific series by its NBER id (e.g. m13045), or browse the ids in one of the 16 subject folders.

Part of Pipeworx — an MCP gateway connecting AI agents to 1683+ live data sources.

Tools

"m13045" or "a01006"). Returns the title, area covered, units, coverage dates and source notes (parsed from NBER's own documentation for that series) plus every dated observation. m = monthly, a = annual, q = quarterly. If you only have the bare digits, it tries monthly, then annual, then quarterly, and returns whichever exists.

• nbermacrohistoryseries(series_id) — one series by its NBER id (e.g.

16 subject-category folders ("01" through "16"). No titles (the directory listing doesn't carry them) — call nbermacrohistoryseries on an id to see what it is.

• nbermacrohistorylist_series(category) — every series id in one of the

Auth

Keyless.

Data sources

archive's landing page (background only, not fetched by this pack).

• — the

the series' raw fixed-width data file. — the series' documentation header (title, area, units, coverage, source, notes).

• —

category's directory listing, scraped for nbermacrohistorylist_series.

• — a

This is a live per-request proxy: every call fetches directly from data.nber.org, nothing is bulk-downloaded or stored by this pack.

Series id → folder. Strip the leading letter (m/a/q); the folder is the first two digits of what remains — "m13045" → folder "13", "a01005a" → folder "01". Verified across categories 01, 07, 08, 11, 13, 15, 16 (2026-09-23). There is no folder 00 or above 16.

data.nber.org 403s any User-Agent containing the literal substring "http", or the substring "bot". Its block page says "Prevent Bots from data downloads." Measured 2026-09-23: "foo/1.0 (+https://x)" → 403, "somebot/1.0" → 403 (on "bot" alone), "foo/1.0 (pipeworx.io)" → 200. Every other pack in this catalog uses a "(+https://pipeworx.io)" User-Agent convention — this one deliberately does not, and cannot, or every request 403s. If you copy this pack's fetch pattern elsewhere against an nber.org/data.nber.org host, keep that in mind.

The per-series .dat file has no missing-data sentinel constant across series — this pack treats any value with abs(value) < 1e-30 as missing, based on the doc header's stated MD=1E-37 convention seen on the series checked so far. If a series' doc header states a different missing-data code, this pack does not read that field — it's a fixed heuristic, not parsed per series.

The .dat files are NOT one row per line uniformly — annual (a-prefix) rows are YEAR VALUE, monthly/quarterly (m/q-prefix) rows are YEAR PERIOD VALUE. This pack tells them apart by counting whitespace-split tokens per line (2 vs 3+), not by trusting the id's prefix letter, since a few ids in the wild (e.g. m13001a) carry a monthly prefix with an annual-shaped suffix.

Shortened. The full README is on GitHub.