@pipeworx/sec-ftd

Connect: https://gateway.pipeworx.io/sec-ftd/mcp · Install: one-click buttons

Tools: 3

SEC fails-to-deliver (CNS) data. Keyless.

Tools

• ftd_security(symbol, periods?) — fails history for one ticker across recent half-month files.
• ftd_largest(period?, limit?) — the biggest fails in one reporting period.
• ftd_periods() — which periods SEC has published (currently 408, back to 2009).

The semantics are the point of this pack

A fail-to-deliver is a settlement failure — shares not delivered by the settlement date. The SEC says, where it publishes this file, that fails are not evidence of naked short selling. Ordinary causes include processing delays, order-entry mistakes, illiquidity in thinly traded securities, and long sales where the seller’s shares did not arrive in time.

There is a loud audience that wants the opposite inference. So the framing is not a footnote: it is in every tool description and rides on every response as what_this_is and not_evidence_of. An agent that reads only the payload still cannot honestly draw the naked-shorting conclusion.

Two more reading traps, both surfaced in the responses themselves:

• The published number is cumulative, not daily-new. It is total shares failing on that settlement date across all participants. Summing days over-counts a fail that persists, which is why ftd_security returns peak_shares_failed alongside the total and says so in reading_note, and why ftd_largest ranks by each security’s worst single day rather than a sum.
• Absence is not a clean bill of health. SEC lists a security on a date only when its fails exceed the reporting threshold, so “no rows” means no reportable fails — not that nothing failed and not that the ticker is unknown. The found: false path says exactly that.

Data source and cadence

https://www.sec.gov/data-research/sec-markets-data/fails-deliver-data — one pipe-delimited file per half month (a = 1st–15th, b = 16th–month end), zipped. Measured 2026-08-25: the current file (cnsfails202607b.zip) is 1.65 MB zipped, 4.6 MB raw, 73,340 rows, covering 13,500 securities. SEC posts a period a few weeks after it closes.

Each call reads the periods the question needs straight from SEC, so currency is simply whatever SEC has published — ftd_periods reports it. A request spanning many periods costs one fetch per half month, which is why periods is capped at 6.

Two upstream traps worth knowing

• The path is not constructible. Recent files live under /files/data/fails-deliver-data/, older ones under /files/data/other/fails-deliver-data/ — verified 2026-08-25, where cnsfails202605b.zip 404s on the first prefix and 200s on the second. The pack reads SEC’s index page rather than guessing.
• These ZIPs set the data-descriptor flag (0x08), so the local file header carries zeroes for both sizes. A reader that trusts the local header extracts nothing. Sizes are taken from the central directory, which always has the real values.

Tools

• ftd_security — Fails-to-deliver history for one US-listed security from the SEC’s CNS fails-to-deliver file, by ticker symbol. Returns each settlement date with the number of shares that failed to deliver, the closi
• ftd_largest — The largest fails-to-deliver in one SEC reporting period, ranked by share count, from the CNS fails-to-deliver file. Returns ticker, security description, settlement date, shares failed and price. A f
• ftd_periods — Which SEC fails-to-deliver reporting periods are currently published, newest first. SEC posts one file per half month (“a” = 1st-15th, “b” = 16th to month end). Use this to pick a valid period argumen