The map is the product: QuantAbundance's public API and MCP server

The standard way to give an AI market context is to bolt a price feed onto it. Prices are the one part a model can get anywhere, and the least useful part on their own: a price without the map around it is a number without a sentence. The part a model cannot get anywhere is the map itself: which names actually belong together, how tightly they really co-move, and why each one is on the board at all.

That map is what QuantAbundance builds, and as of this week the whole of it is queryable by machine: a public JSON API at api.quantabundance.com, and an MCP server that plugs the corpus straight into Claude or any MCP-capable assistant. No key, no signup, free at the point of use.

What is actually in the corpus

The bubble map, measured rather than asserted. Twelve editorial clusters of the AI supercycle (compute, memory, power, photonics, robotics, quantum and the rest), each tested against 252-day residualized correlations. The taxonomy is honest about its own failures: clusters that turn out to be SPY in a costume are labeled broken, with the rationale attached. The methodology and the ranking live in the 12 bubbles, ranked.

A thesis per instrument, ~700 of them. Not a description scraped from a filing: a written position on what the company is in the value chain, who anchors its revenue, what backlog is disclosed, and which cluster it trades with. This is the layer that is not derivable from any price feed at any price.

Correlations that answer a specific question. Whether two names actually move together, or only share a story. Backward-looking, 252-day window, residualized: the API says what was measured, not what happens next.

The 24/7 tape, from the one source we can legally hand you. Around sixty QA instruments carry a liquid Hyperliquid perp (the live count is whatever /hyperliquid/markets returns, and it moves as the venue lists and delists), and the API serves each one's live mark, hourly funding, open interest and 24h volume, continuously, while the primary exchanges are shut roughly 80% of the week. Licensed exchange OHLCV is exactly what a research desk may NOT redistribute, so the public API serves price summaries there instead of pretending otherwise; the Hyperliquid lane is redistributable in full, and it is the same feed the /hyperliquid board renders. Trading the perps themselves is a different activity with its own risk profile: /stack/hyperliquid.

The MCP server: the corpus inside your assistant

MCP is the open protocol that lets an AI assistant call external tools. QA's server exposes seven: list and inspect bubbles, list themes, pull an instrument's thesis, resolve a name to a ticker, check a correlation, list tracked 13F portfolios. One command wires it into Claude Code:

claude mcp add quantabundancia -- uvx --from "git+https://github.com/LXunchained/QuantAbundancia#subdirectory=mcp_server" quantabundancia-mcp

The design detail that matters for anyone who has piped an API into a model before: responses are shaped for a context window, not for a browser. The raw correlation matrix is ~200KB of JSON (86 tickers, 3,321 pairs); the tool returns a bounded view, ~19KB unfiltered and ~8KB filtered to a single ticker, sorted strongest-relationship-first so the top of the list is the part worth reading. Whenever a cap truncates a list the response says so, in words ("Showing 120 of 161. Narrow the request to see the rest."), because a silently shortened list reads to a model as the complete set.

What that turns into in practice: ask an assistant "does $NVDA actually trade with the memory names or just share their story", and instead of hallucinating a number it pulls the measured 252-day answer, the bubble membership on both sides, and the written thesis for each name involved.

Access, and where this is going

No key at all: 200 requests a day, every endpoint and every MCP tool, zero configuration. A free key lifts that to 1,000 a day. Both stay free; the free lane is not a trial.

For sustained programmatic use, the grid, now on the page instead of "email us and guess":

Need more than Desk, or an unmetered feed inside an internal tool? That is a conversation, not a checkout: [email protected].

The price is set; the plumbing is honest about where it is. Keys are issued on request at that address while self-serve checkout is finished, and early users keep preferential terms, the same logic as the Founding tier: the people who show up before it is automated keep the better deal. What the number is no longer depends on you guessing it.

What to watch

• Coverage of the Hyperliquid lane: the live count is served by /hyperliquid/markets and moves in both directions as the venue lists and delists equity perps.
• Whether the correlation window grows alternatives (shorter windows for regime detection is the most-requested shape so far).
• Self-serve checkout: the grid is published (above), so the remaining step is keys minted at checkout instead of issued by email.
• MCP protocol adoption: every assistant that speaks it becomes a client of this corpus with zero integration work on our side.

Try it now: the board at /hyperliquid, the map at /bubbles, the server at github.com/LXunchained/QuantAbundancia.

The research side: higher assistant limits and operator commentary are part of /pro.

QuantAbundance is educational research. Nothing here is investment advice. See /disclosures.