Track: Real World Assets (Build with CMC: API Hackathon)
Tokenised Nvidia trades as
NVDAX,NVDAon,NVDAB,rNVDA, and four more wrappers — each with its own price, liquidity and issuer. Parity measures every wrapper against the real instrument and against every rival wrapper, so you can see the premium before you pay it.
A tokenised share is the same economic claim sold through different issuers. The wrapper — not the asset — sets the price, and nothing on the market compares them:
| Observed dislocation | Source |
|---|---|
| A tokenised AAPL traded at a +14.92% premium with $195 of depth before 2% slippage | Ondo public post-mortem of early tokenised equities |
| The same SpaceX share priced $122 → $176 across venues within ~10 days of listing | CMC Research, Tokenized Stocks: The Venue Landscape |
NVDAon traded +8.83% above the underlying on 6 Sep 2025 while NVDAX sat in line |
Parity's own history engine, shown in-app |
SPYon priced +83 bps above the cheapest SPY wrapper at the time of writing |
Parity live board |
Every one of these is a real cost to a buyer, and every one is invisible on an issuer's own dashboard.
- Prices the asset — blended token price versus the underlying's own last print, with the session and its age always shown.
- Ranks every wrapper — cheapest to dearest, with unit normalisation, liquidity gating and outlier detection. Excluded venues stay visible with the reason attached.
- Proves the history — 180 sessions of drift-free cross-wrapper spread plus premium-versus-underlying, with weekend/holiday bands shaded. The RWA API has no history endpoint yet, so Parity reconstructs it from token OHLCV joined against the underlying.
- Alerts on your thresholds — evaluated server-side against live data on every page load; the watchlist lives in your browser, so there are no accounts and no database.
- Shows its work — every verdict links to the raw CoinMarketCap call behind it: endpoint, HTTP status, credits burned, full JSON payload.
| Wrapper verdict | Dislocation history | Evidence drawer |
|---|---|---|
![]() |
![]() |
![]() |
Request (the key is supplied via header and never committed):
curl -s "https://pro-api.coinmarketcap.com/v5/real-world-assets/quotes/latest?symbol=NVDA&convert=USD" \
--header "X-CMC_PRO_API_KEY: $CMC_API_KEY"Trimmed live response — note tokens[]: each wrapper carries its own price, volume and issuer, which is what makes cross-issuer comparison possible:
{
"data": {
"rwa_assets": [{
"name": "Nvidia Corp",
"symbol": "NVDA",
"rwa_id": 2,
"asset_type": "stock",
"average_tokenized_price": 213.7948,
"tokenized_volume_24h": 83823811.7,
"tokens": [
{ "symbol": "NVDAX", "crypto_id": 36992, "price": 214.036, "volume_24h": 12932112.23, "issuer_name": "Backed Assets" },
{ "symbol": "NVDA.D", "crypto_id": 28616, "price": null, "volume_24h": null, "issuer_name": "Dinari Assets" },
{ "symbol": "NVDAon", "crypto_id": 38093, "price": 214.040, "volume_24h": 2410487.34, "issuer_name": "Ondo Assets" }
]
}]
},
"status": { "timestamp": "2026-09-16T07:41:31.545Z", "error_code": "0", "credit_count": 1 }
}The same request/response pair is inspectable in the running app via the evidence drawer on any asset page (/asset/NVDA).
| Endpoint | What it contributes | Cost |
|---|---|---|
/v5/real-world-assets/quotes/latest |
Core call. Every wrapper's own price, volume, market cap and issuer, plus CMC's blended tokenized quote | 1 credit / 250 assets |
/v5/real-world-assets/assets/list |
The investable universe (~7,940 assets) ranked by tokenized volume | 1 credit / 250 assets |
/v5/real-world-assets/map |
Stable rwa_id resolution for search and evidence pinning |
free |
/v5/real-world-assets/info |
Industry, employees, founding date, SEC CIK, description | 1 credit / 250 assets |
/v5/real-world-assets/issuers/list |
Issuer lineage behind each wrapper (Backed, Ondo, bStocks, Paxos…) | 1 credit / request |
/v2/cryptocurrency/ohlcv/historical |
Daily wrapper closes → premium and spread history | 1 credit / ~100 points |
A full-universe dislocation scan costs 2 credits. See /endpoints in the app for the live call log.
The per-issuer split in quotes/latest is the single design decision that makes Parity possible; a whole-universe screener for two credits is remarkable value. Eight specific frictions are documented in-app at /feedback, with reproductions:
market-pairs/listis documented for Startup but returnserror_code 1006on the hackathon key — the one endpoint that would add per-venue depth.- OHLCV
historicalreturns exactly 10 points whentime_startis used withoutcount— a range query that behaves like a paginated one. - No RWA historical series yet (the academy article calls it "planned") — every builder must reconstruct premium history themselves.
- No token-economics metadata, so total-return accrual reads as mispricing: Ondo's
*onwrappers legitimately sit above the ticker, and a naive cross-issuer comparison reported 263 bps of Ford dividends as an arbitrage opportunity. - No session alignment: RWA candles close 23:59:59 UTC while US equities close 20:00/21:00 UTC, so naive comparison bakes after-hours drift into "premium".
- No unit metadata on commodities — gold is quoted per troy ounce and per gram in the same family.
- Zero-volume venues still carry prices (one AAPL token sits −9.5% from its family on $0.00 of 24h volume) with nothing to distinguish them from live ones.
- Derivatives and wrapped tokens sit in the same
tokens[]array as primary issuance; atoken_classfield would remove the string-matching.
Eligibility pipeline, in order: presence → unit → liquidity → outlier, then verdict maths on what survives.
blended_price = median(included wrapper prices)— median, so one bad venue cannot move the headlinepremium % = (blended_price / reference_price − 1) × 100wrapper_spread_bps = (dearest / cheapest − 1) × 10,000— cheapest must clear a volume bar scaled to market size- Verdicts:
OVERPRICED(> +0.75%),DISCOUNT(< −0.75%),FAIR(±0.75%),DISPERSION(cross-wrapper stdev > 1.5%),ILLIQUID(< $50K volume),RELATIVE(no mapped reference) - Per-gram metal quotes are normalised ×31.1035; anything with a genuinely different unit is excluded and explained rather than reported as a fake discount
- Reference prices come from public exchange endpoints with a second provider as failover, always labelled with session and age; futures/ADR/leveraged references cap confidence
Full detail: /methodology in the app. The engine is a pure function with 18 unit tests covering unit mismatches, dead venues, outlier rejection, staleness and every verdict branch.
- Next.js 16 (App Router) + TypeScript + Tailwind, no chart library and no client data fetching for prices
- All CoinMarketCap calls happen server-side through one client that handles caching (per-endpoint TTLs that respect CMC's own refresh cadence), retry with backoff, in-flight de-duplication, and evidence capture
- The API key never leaves the server: not in client bundles, logs, cache, evidence records or the repo. Evidence stores the URL plus a curl that reads
$CMC_API_KEYfrom the environment, and.env*is gitignored and.vercelignored - Rate-limit aware: a request gate caps concurrent reference fetches at 2 with 250ms spacing, so bursts never trip provider throttling
pnpm install
cp .env.example .env.local # add your CMC Pro API key
pnpm dev # http://localhost:3000
pnpm test # verdict engine unit tests
pnpm lint && npx tsc --noEmit # checksGet a free key at coinmarketcap.com/api — the RWA endpoints are available from the Basic plan.
Parity exposes the same engine as five MCP tools over Streamable HTTP, with no auth and no sessions:
| Tool | What it answers |
|---|---|
parity_fair_value |
Blended premium, cheapest wrapper, reference freshness, confidence |
parity_spread |
Every wrapper ranked on accrual-adjusted prices, with exclusion reasons |
parity_screener |
Top cross-wrapper dislocations across the universe |
parity_history |
Weekend vs weekday dislocation, worst episodes, recent spread series |
parity_search |
Find assets by ticker, name or slug |
claude mcp add --transport http parity https://parity-xi.vercel.app/api/mcpThen ask “is tokenised NVDA rich right now?” or “which SPY wrapper is cheapest
after dividends?”. Try every tool live, in the browser, at
/mcp — the tester performs the real MCP
handshake and shows the JSON-RPC wire traffic. The CoinMarketCap key never
leaves the server; agents see the same cached, accrual-adjusted numbers as the
web app.
Known limitations (stated, not hidden)
- Total-return wrappers are accrual-adjusted, not excused. Ondo tokens reinvest dividends into the token, so Parity reconstructs each token's reinvestment factor from the underlying's dividend events (measured from the token's own inception) and compares wrappers on
token_price / factor. The raw price, the adjusted figure and the accrued percentage stay visible, and the verdict states the factor and inception date. Cross-model spreads that survive the adjustment (e.g. GameStop's ~200 bps on a zero-dividend name) are the genuine dislocations. - Reference prices are not from CoinMarketCap. The RWA API does not expose the underlying instrument's own quote, so Parity uses public exchange endpoints with failover; when both fail it degrades to relative mode instead of guessing.
- History starts when CMC's OHLCV coverage starts, not at token launch.
- Venue depth stops at issuer level while
market-pairsis unreachable on this plan.
- Track: Real World Assets
- Deployed: https://parity-xi.vercel.app
- [Demo video](https://youtu.be/8oL0CkcFJFk)
- API feedback:
/feedback(seven reproducible findings)
Built for the Build with CMC: API Hackathon. Market data © CoinMarketCap. Not investment advice.



