Sec Nport MCP server

@pipeworx/sec-nport

Mutual fund and ETF portfolio holdings, from SEC Form N-PORT. Keyless.

Part of Pipeworx — an MCP gateway connecting AI agents to 1576+ live data sources.

Tools

• nportfundowners(security, quarter?, limit?) — which funds and ETFs hold a stock or bond, ranked by position size. Takes a ticker, a company name, or a CUSIP.
• nportfundportfolio(fund, quarter?, limit?) — one fund's reported positions, largest first.
• nportholdingscoverage() — which releases are loaded, how current they are, and what the data does not cover.

The inversion is the point of this pack

"Which ETFs hold the largest positions in NVDA?" returned nomatch before this shipped, and the router explained the gap in its own words: it would have to check each ETF individually. edgarfund_holdings goes fund → holdings. Nothing went security → funds, because SEC publishes N-PORT filing-first and so does every API over it. That is a shape problem, not a rate-limit problem.

This is not 13F warmed over. 13F is institutional managers over $100M, long US equity only. N-PORT is registered funds — mutual funds and ETFs — and carries their bond positions too. "Which funds hold this corporate bond" has no 13F answer at all.

Scope, stated because a partial load read as a complete one is a wrong answer

The quarterly archive is 441MB across 32 tables. Loaded: equity, preferred and debt positions — about 3.6M of the 5.35M holdings rows per quarter, across 14,416 funds. Not loaded: derivatives, loans, repos and structured products. A swap notional is not a holding of the underlying, and reporting it as one would be a wrong answer wearing a right one's clothes. Every response carries this in its scope field.

Two more caveats ride on every answer:

quarter ends. Responses carry asofperiod. An answer that implies real-time is wrong.

• N-PORT is filed monthly but released publicly on a quarterly lag, roughly 60 days after the

the series name is the only hint. An ETF-only answer cannot be given honestly from this filing, so the pack says that rather than inventing the classification.

• N-PORT has no ETF flag. These are registered funds, which include ETFs and mutual funds alike;

Where the data lives, and why it is not in R2

All of it is in Postgres. Measured on the live database, one release loaded:

That is a real amount of Postgres for one pack, so the choice is worth stating rather than leaving for someone to rediscover at 3am.

The query is the reason. The inversion this pack exists for — which funds hold this security — is a selective lookup on cusip and issuer_name returning tens of rows out of 3.6M. That is precisely what a B-tree index is for: it answers in milliseconds and touches almost none of the 1.1GB. R2 has no indexes. Serving the same question from object storage means either scanning the objects or building a side index to point into them, which is a worse database than the one already here.

R2 earns its place when the access pattern is bulk: whole-file reads, archival quarters nobody queries interactively, or payloads too large to index usefully. A reasonable split later is hot quarters in Postgres, cold quarters in R2 — but that only pays once several quarters are loaded, and today exactly one is. Splitting now would add a storage tier to save nothing.

Where the line is, so the next person does not have to guess. Each quarter costs roughly 1.1GB. Around four to six loaded quarters this becomes the largest single consumer in the database and the cold-quarter split should be built rather than discussed. sec13fholdings sits at 3.8M rows for comparison — this pack is the same order of magnitude, not a new class of problem.

Ticker resolution

Shortened. The full README is on GitHub.