The public record, as infrastructure.

What Congress trades, what insiders file, who wins federal contracts, what the Fed said: LuxAlgo Market Trackers turns the public record of US markets into one clean dataset you can query, subscribe to, or hand to an AI agent. Every row links back to the government document it came from.

LuxAlgo Market Trackers is a LuxAlgo open-source project. The official repository is github.com/LuxAlgo/market-trackers, and the published data lives at github.com/LuxAlgo/market-trackers-data.

---

LuxAlgo Market Trackers ingests 18 datasets from primary sources only, normalizes them into one schema, stores them in a local database, serves them to AI agents over MCP, and republishes everything as free daily data dumps. Every row carries provenance: a working deep link to the primary document it was parsed from. No accounts, no telemetry, and every source is keyless except patents (a free USPTO Open Data Portal key).

Quickstart

The only identity LuxAlgo Market Trackers ever sends is a contact email inside the User-Agent that the SEC's fair-access policy requires for EDGAR.

git clone https://github.com/LuxAlgo/market-trackers && cd market-trackers
pnpm install && pnpm build
alias market-trackers="node packages/cli/dist/index.js"

# 1. Pull data. Short-sale volume needs nothing at all:
market-trackers sync --source finra

# 2. Insider filings need the SEC-required contact email:
market-trackers sync --source edgar --contact you@example.com --limit 500

# 3. See what you have:
market-trackers status

# 4. Serve it to your AI agent over MCP:
market-trackers serve

The same commands ship on npm as npx @luxalgo/market-trackers-cli … when you'd rather not hold a checkout.

Or skip ingestion entirely: the published dumps are a rebuildable archive, and market-trackers import is the exact mirror image of market-trackers export:

git clone https://github.com/LuxAlgo/market-trackers-data
market-trackers import ./market-trackers-data

Everything is incremental and idempotent: market-trackers sync picks up where it left off (per-source watermarks) and re-running never duplicates a row (natural-key upserts). SQLite by default; --db postgres://… is the same store behind one flag.

Datasets

Dataset	Primary source	What you get	Status in this build
Congress trades	Senate eFD + House Clerk financial disclosures	Every reported transaction, amounts as disclosed ranges, member identities	✅ ingesting
Insider transactions	SEC EDGAR Forms 3/4/5 (XML + quarterly sets)	Every insider transaction and holding row, raw SEC codes + legend	✅ ingesting
13F holdings	SEC EDGAR 13F-HR information tables	Quarterly institutional holdings, CUSIP-keyed, values normalized to dollars	✅ ingesting
Government contracts	USAspending API	Federal contract awards with best-effort recipient→ticker mapping	✅ ingesting
Government grants	USAspending API	Federal grant awards, through the same conservative recipient→ticker mapping	✅ ingesting
Lobbying	Senate LDA API	Lobbying filings with client→ticker mapping	✅ ingesting
Short-sale volume	FINRA Reg SHO daily files	Symbol-day short volume + short ratio, both pre- and post-2026 file formats	✅ ingesting
Committee assignments	unitedstates/congress-legislators	Which member sits on which committee, the "oversees what they trade" join	✅ ingesting
Patents	PatentsView bulk data (USPTO Open Data Portal)	Granted patents by assignee, joined to tickers	✅ ingesting (free key)
Clinical trials	ClinicalTrials.gov API v2	Study registrations and status by sponsor, joined to tickers	✅ ingesting
FDA drug approvals	openFDA (Drugs@FDA)	Approval actions as the FDA recorded them: history, never a prediction	✅ ingesting
Futures positioning (COT)	CFTC Commitments of Traders	Weekly legacy-report positioning per market	✅ ingesting
Federal legislation	GPO GovInfo bulk BILLSTATUS	Bill titles, sponsors, latest actions, joined to who lobbies on each bill	✅ ingesting
Campaign finance: totals	FEC bulk downloads	Candidate-cycle receipts/disbursements/cash, verbatim FEC numbers	✅ ingesting
Campaign finance: PAC→candidate	FEC bulk downloads	Committee→candidate contributions as filed (refunds stay negative)	✅ ingesting
Congressional hearings	GPO GovInfo CHRG (sitemaps + MODS)	Hearing transcripts indexed: title, committees, witnesses, member ids, deep links	✅ ingesting
Fed communications	Federal Reserve Board public JSON feeds	FOMC statements/minutes announcements, speeches, testimony, as an index with links	✅ ingesting
Wikipedia pageviews	Wikimedia REST pageviews API	Daily article views for a curated public-company map: attention, measured	✅ ingesting

market-trackers status always tells you exactly what your build ingests; sources that aren't implemented yet say so out loud instead of pretending.

Ask your agent, get receipts

market-trackers-mcp serves the whole store to any MCP client, locally over stdio or hosted over stateless streamable HTTP. Every answer carries primary-source citations: when an agent tells you what Congress bought this week, it can show you the actual filings.

// Claude Desktop / Claude Code / any MCP client
{
  "mcpServers": {
    "market-trackers": {
      "command": "npx",
      "args": ["-y", "@luxalgo/market-trackers-mcp"],
      "env": { "MARKET_TRACKERS_DB": "/path/to/market-trackers.db" },
    },
  },
}

Tool	What it answers
tracker_congress_trades	"What did Congress buy this week?", with per-filing citations
tracker_congress_members	Resolve member names; who trades most
tracker_member_profile	One member's whole public footprint: trades, committees, top tickers
tracker_committees	Committee rosters: who sits where, joined to what they trade
tracker_insider_trades	Form 3/4/5 rows by ticker/insider/code/value, with the SEC code legend
tracker_insider_summary	Buys vs sells, net shares, notable insiders for a ticker (arithmetic only)
tracker_13f_holders	Who holds a security, with share changes vs the prior quarter
tracker_13f_manager	One manager's holdings ("what does Berkshire hold?")
tracker_gov_contracts	Federal contract awards by ticker/recipient/agency
tracker_gov_grants	Federal grant awards by ticker/recipient/agency
tracker_gov_contract_totals	Government revenue over time: award counts + sums per year/quarter
tracker_lobbying	Lobbying filings by ticker/client/registrant
tracker_short_volume	Daily short-sale volume series for a ticker
tracker_patents	Granted patents by ticker/assignee
tracker_clinical_trials	Studies by ticker/sponsor/status/phase
tracker_fda_approvals	Drugs@FDA approval actions by ticker/company/drug
tracker_cot	Weekly futures positioning series for a market
tracker_bills	Bill status by title/sponsor/policy area, plus who lobbies on that bill
tracker_campaign_finance	Candidate money: FEC totals + itemized committee→candidate contributions
tracker_congress_hearings	Hearing transcripts by committee/witness/chamber: "who testified about X?"
tracker_fed_communications	FOMC statements, minutes, speeches, testimony: what the Fed published, linked
tracker_wiki_pageviews	Wikipedia attention series for a company, by ticker or article
tracker_search	One query across tickers, members, managers, insiders
tracker_freshness	How old every dataset is, so agents can check before they answer

tracker_freshness is first-class on purpose: an agent quoting stale congress data without knowing it is the failure mode that kills trust. And the committee tools exist because the most-asked question about congressional trading, "do they trade what they oversee?", is a join, not a score: LuxAlgo Market Trackers gives you both sides of it with receipts and leaves the conclusion to you.

The dumps

market-trackers export writes the full store as versioned JSON: daily deltas per dataset, full snapshots sharded by event year, an RSS feed per dataset, and a manifest with row counts, watermarks, and per-source health. A scheduled CI workflow publishes them to a public data repository (LuxAlgo/market-trackers-data) so analysts and journalists who will never run code still get the data: one URL, no signup.

congress/trades/2026/2026-08-24.json      # daily delta: what was ingested that day
congress/trades/latest.json               # newest delta, stable URL
congress/trades/feed.xml                  # RSS 2.0 over the newest rows
congress/trades/feeds/by-member/P000197.xml   # per-member feed: follow ONE member
congress/trades/feeds/by-ticker/NVDA.xml      # per-ticker feed: follow ONE stock
congress/trades/snapshot-2026.json.gz     # full history, sharded by event year
congress/trades/snapshot-2026.parquet     # …with a Parquet sibling for columnar tooling
manifest.json                             # row counts, freshness, source health
explorer/index.html                       # zero-build data browser (works on GitHub Pages)

The two most time-sensitive datasets (congress trades, insider transactions) also get an intraday fast lane that tops up their deltas between daily publishes, and a weekly workflow mirrors the whole data repo to a Hugging Face dataset for datasets.load_dataset(...) consumers. The full contract is in docs/market-trackers-data.md.

Code is MIT; the dumps are CC0: public-domain-derived government data stays public domain. See data-licenses/.

Alerts without a server

Every dataset directory carries a feed.xml: plain RSS 2.0 over the newest ingested rows, each item a strictly factual one-line restatement of a row whose link goes straight to the primary-source document. Point a feed reader, a chat integration, or a five-line script at one URL and you have congress-trade alerts with receipts. No server, no webhooks, no account, nothing to deploy.

It's also per-entity: feeds/by-ticker/{TICKER}.xml under every ticker-bearing dataset, and feeds/by-member/{bioguideId}.xml under congress trades. One URL follows one stock's insider filings, or one member's disclosures, and nothing else. Feeds exist for entities active in the last 30 days (capped at the 200 most active per dataset; the manifest carries the exact counts, so the cap is never silent).

Deep history

market-trackers backfill --source edgar --from 2004-01-01 walks a source as far back as its free history goes: chunked, resumable, watermark-driven, so an interrupted run picks up exactly where it stopped instead of re-walking covered ground. The backfill.yml workflow runs the same engine on CI and publishes the resulting year-sharded snapshots as release assets on the data repo, so nobody has to re-crawl a decade of filings that CI already crawled. And because market-trackers import rebuilds a store from published dumps, the archive is not a pile of downloads; it's a restorable database. Per-source depth notes: docs/backfill.md.

Bring your own prices

market-trackers analyze congress --prices your-prices.csv joins disclosed trades against a price series you supply (date,ticker,close) and reports the price change over a window after each disclosure: the classic "what happened after they filed?" table. LuxAlgo Market Trackers ships no market data and computes no scores; the output is arithmetic between public records and your own inputs, every result carries a disclaimer saying exactly that, disclosed amounts stay ranges, and the timeline anchors on the filing date, because that's when the public actually learned.

market-trackers backtest congress --prices your-prices.csv --member <name> --window 30 goes one step further: one fixed, fully disclosed strategy (enter at the first close after the disclosure, exit after the window, equal weight per event) with win rate, mean/median change, and every skipped event kept in the output with its reason. No parameter fitting, no optimization, no costs modeled, and disclosed amount ranges are never used as position sizes. The point is an honest baseline you can reproduce, not a pitch. Details: docs/analytics.md.

On a chart

The rows chart well, and the hosted trackers show how: the insider tracker on LuxAlgo paints every Form 4 fill onto the ticker's price tape with Vela, LuxAlgo's open-source charting engine (@luxalgo/vela, Apache-2.0), as a custom renderer layer: the same tracker_insider_trades rows this repo serves, each marker still linking to its filing. That happens in the visitor's browser. Nothing here draws or serves a chart, and the price bars come from the page's own market-data source, never from LuxAlgo Market Trackers, which ships no market data by design. To do the same with your own store, take the rows from the MCP tools or the dumps, supply bars via Vela's data option or a provider, and paint the markers through Vela's plugin SDK.

Python

The dumps are plain JSON with Parquet siblings, so nothing about LuxAlgo Market Trackers requires JavaScript. The dependency-light reader in python/ (pandas optional) loads them from a URL or a checkout:

from market_trackers_data import load_snapshot, to_dataframe

rows = load_snapshot("https://raw.githubusercontent.com/LuxAlgo/market-trackers-data/main", "congress-trades")
df = to_dataframe(rows)  # plain list[dict] if pandas isn't installed

Worked, reproducible examples live in notebooks/: price change after disclosure (bring-your-own-prices) and the committee-oversight join.

Data health

Honesty is the point: pipelines rot, formats drift, and a data project that can't show you its health isn't one you should trust. Daily canaries fetch each source, assert parse success rates, and fingerprint page/file formats so drift turns the board red before a parser silently misparses. This board is generated from the latest canary run:

Source	Status	Last checked
SEC EDGAR (Forms 3/4/5, 13F-HR)	🟢 healthy	2026-09-07T15:22:46.742Z
edgar-bulk	🟢 healthy	2026-09-07T15:22:46.867Z
Senate eFD (PTRs)	🟡 stale	2026-09-07T15:22:49.000Z
House Clerk (PTRs)	🟡 stale	2026-09-07T15:22:49.360Z
USAspending (contracts + grants)	🟢 healthy	2026-09-07T15:22:50.168Z
Senate LDA	🟢 healthy	2026-09-07T15:22:51.151Z
FINRA Reg SHO	🟢 healthy	2026-09-07T15:22:51.412Z
Committee assignments	🟢 healthy	2026-09-07T15:22:51.900Z
PatentsView (patents)	🟢 healthy	2026-09-07T15:22:52.074Z
ClinicalTrials.gov	🟢 healthy	2026-09-07T15:22:52.132Z
openFDA (Drugs@FDA)	🟢 healthy	2026-09-07T15:22:52.635Z
CFTC COT	🟡 stale	2026-09-07T15:22:53.007Z
Wikimedia pageviews	🟢 healthy	2026-09-07T15:22:53.071Z
GovInfo (bill status)	🟢 healthy	2026-09-07T15:22:56.509Z
FEC campaign finance	🟢 healthy	2026-09-07T15:22:58.433Z
GovInfo CHRG (hearings)	🟢 healthy	2026-09-07T15:22:58.461Z
Federal Reserve (communications)	🟢 healthy	2026-09-07T15:22:58.583Z

Principles

• MIT code, CC0 data. The public record belongs to the public, including in machine-readable form.
• Primary sources only. SEC EDGAR, Senate eFD, the House Clerk, USAspending, the Senate LDA API, FINRA, the unitedstates/congress-legislators project, PatentsView bulk data via the USPTO Open Data Portal, ClinicalTrials.gov, openFDA, the CFTC, GPO GovInfo (bills and the CHRG hearings collection), the FEC, the Federal Reserve Board's public feeds, and the Wikimedia pageviews API. Never scraped from commercial products.
• Receipts everywhere. Every row carries provenance: the primary-source URL, retrieval time, parser identity, and a confidence tier. Rows extracted from scanned documents are flagged needsReview instead of silently trusted.
• No telemetry. LuxAlgo Market Trackers never phones home. The only outbound identity is the SEC-required contact email in the EDGAR User-Agent, and it goes to the SEC, nobody else.
• Fair access, enforced in code. The EDGAR client is hard-capped at 10 requests/second by a rate limiter with a unit test, not by a sentence in a README.

Non-goals

Things LuxAlgo Market Trackers deliberately does not do:

• No predictions, signals, scores, or "conviction" ratings. LuxAlgo Market Trackers is data with receipts, not advice. shortRatio, buy/sell counts, and market-trackers analyze's price changes are arithmetic, not recommendations.
• No fabricated precision. Congressional amounts are disclosed as ranges; LuxAlgo Market Trackers stores the range bounds and never invents midpoints, not even in analytics output.
• No social/Reddit sentiment. Redistribution restrictions make it unshippable in dumps; it doesn't belong here.
• No scraping of commercial data products. If it isn't a primary source, it isn't in LuxAlgo Market Trackers.
• No editorially curated calendars presented as public record. LuxAlgo Market Trackers ships the FDA's own record of approval actions (Drugs@FDA); it does not ship hand-maintained future-decision-date calendars or any other dataset that needs human curation to exist.
• No data whose license forbids open redistribution. Free-to-view is not free-to-republish: FINRA's OTC/ATS transparency product, for example, stays out of the CC0 dumps because its terms don't permit it; the investigation and decision are documented in docs/decisions/0004-ats-otc-data-licensing.md.

Architecture

packages/
  core/   @luxalgo/market-trackers-core   ingestors, parsers, zod schemas (source of truth),
                                   storage (SQLite default / Postgres via one flag),
                                   entity resolution, backfill, dump export/import,
                                   BYO-prices analytics, canaries
  mcp/    @luxalgo/market-trackers-mcp    MCP server: stdio + stateless streamable HTTP
  cli/    @luxalgo/market-trackers-cli    market-trackers sync | status | export | import | backfill |
                                   resolve | analyze | backtest | canary | serve
python/   market-trackers-data             dependency-light reader for the published dumps
notebooks/                         worked examples over the dumps (no keys, reproducible)
templates/market-trackers-data/            the data repo's README, LICENSE, and static explorer

Design decisions are documented in docs/decisions/, per-source implementation notes in docs/sources/, operations in docs/operations.md, and the golden-file testing policy in CONTRIBUTING.md.

License

Code: MIT © LuxAlgo Global, LLC. Published data dumps: CC0. The LuxAlgo name and logo are trademarks (see TRADEMARKS.md); forks fly their own flag. Vulnerabilities: SECURITY.md.

Everything here is descriptive arithmetic over public records. Nothing in this repository or its published data is investment advice, and no predictive claim is made.