Spiral Stake β mcp read-data backbone (Step 1a)
The composition authority + read API for Spiral Stake strategy data. It ports the app's
client-side data composition server-side and serves it back as JSON, so three consumers run
off one core:
- The app β the complete read surface it currently fetches, for a like-for-like Step-2 swap.
- Agents β the frozen
/strategies contract.
- Partners (later) β a clean, versioned, self-documenting REST surface (OpenAPI).
Read-only. No contract writes, no tx building, no signing, no private keys. It composes
numbers and serves JSON. The MCP tool layer, execution/build_*, and partner auth are later
steps β the seams are pre-cut (core + http) but not built.
Architecture
src/
core/ pure composition β no I/O, unit-testable
leverage.ts β VERBATIM copy of v2-client/src/utils/leverage.ts (byte-diff gated)
apy.ts collateral-APY resolution (Pendle/DeFiLlama/Royco/StableWatch/on-chain)
leverageApy.ts 30/60/90d leveraged-APY windows + lumpy-token smoothing
compose.ts Market assembly from warm raw (port of FlashLeverage.createInstance)
strategy.ts Market β frozen /strategies contract (ladder, freshness, spiralHints)
exitLiquidity.ts exit-liquidity tier (the single verdict) + thresholds
sources/ one adapter per upstream (fetch + parse). native fetch + viem; no axios/wagmi
cache/ two-layer cache β RAW store (last-good) + warming policy (CONTRACT cadences)
warmer/ background scheduler (stale-while-revalidate); reads NEVER fetch
http/ thin hono REST layer (the only thing that knows about requests)
data/ static market config (copied from the app) + readMarkets port
config/ env (validated) + structured logger with correlation ids
Data flow: warmer fetches upstreams on the CONTRACT cadence β writes raw to the cache
(last-good on failure) β core composes strategies from warm raw only β http serves them.
A stale/failed upstream serves last-good with a visible stale age per field-group β never
dropped, never coerced to 0.
Run
cp .env.example .env
npm install
npm run dev
npm test
npm run typecheck
npm run openapi
node --import tsx scripts/smoke.ts
/health (liveness + cache diagnostics) and /ready (503 until the warmer primes the required
data) are available immediately on boot.
Endpoints β app-read inventory β endpoint mapping
Every data read the app performs in v2-client/src/api-services/* maps to an endpoint here
(positions stay on the dashboard backend and are not reimplemented):
App read (api-services/*) | mcp endpoint |
|---|
getTokenApy, getAllMorphoMarketsData, fetchMerklSpotIncentives, prices, on-chain value β composed markets | GET /v1/strategies, GET /v1/strategies/:id |
FlashLeverage.createInstance β the app's full client-side Market[] composition | GET /v1/app/markets (raw domain model; BigNumberβ{"$bn":β¦}, bigintβ{"$bigint":β¦} β revive before use) |
morpho.ts getAllBorrowApyHistories / getBorrowApyHistory | GET /v1/markets/borrow-apy-history, GET /v1/markets/:id/borrow-apy-history |
apy.ts resolveTokenApyHistory (DeFiLlama/Royco/StableWatch history), chart.ts getApyChart | GET /v1/collateral/apy-history, GET /v1/collateral/:id/apy-history |
merkl.ts fetchMerklIncentiveData (borrow-incentive APR history) | GET /v1/markets/:id/incentive-history |
token.ts getAllLoanTokenPrices / getTokenPrice | GET /v1/prices |
chart.ts getMarketChart (CoinGecko price chart) | GET /v1/prices/chart?coinId=&days=¤cy= (on-demand proxy) |
dashboard.ts getApySnapshot (StableWatch stable APY) | GET /v1/stable-apy β fetched directly from StableWatch as a warmed upstream; the mcp owns this data now and no longer depends on the dashboard /apy endpoint |
| swap/meta-dex aggregators, referral, positions | out of scope (execution / backend domains) |
Cross-cutting: /v1 versioning, CORS limited to CORS_ORIGINS, a consistent error envelope
({ error: { code, message, correlationId } }), a correlation id per request (echoed as
x-correlation-id), structured JSON logs, and an OpenAPI 3.1 spec at /openapi.json (+ committed
openapi.json).
Parity gates
leverage.ts golden-vector β test/golden/leverage.golden.test.ts freezes fixed inputs β
exact outputs; leverage.parity.test.ts asserts the file is a byte-for-byte copy of the
app's leverage.ts. This locks every APY/LTV/ladder number to the app.
/strategies composition β test/parity/strategies.compose.test.ts seeds the raw cache
with a controlled fixture and asserts the composed contract field-by-field (sourcing, LTV math,
ladder via the verbatim leverage.ts, freshness cadences, null-vs-absent, spiralHints
isolation). Runs offline in CI.
- Live golden-set + app-surface parity β
scripts/capture-parity.ts primes the warmer
against real upstreams and writes the composed /strategies (+ apy/borrow histories) for a
fixed market set to test/fixtures/. The parity diff vs the app is produced by feeding the
same fixed inputs to the app's client-side composition and comparing (the app logs its composed
markets); the PR commits the fixtures and the empty diff. This is the gate that lets us offload
TVL safely.
Frozen decisions (parity-critical)
bignumber.js pinned to the app's 9.3.1; no global BigNumber.config change β
toFixed(2) / rounding (ROUND_HALF_UP) identical.
- On-chain reads via viem replace wagmi:
getCollateralValueInLoanToken (multicall3), stUSDS
str(), spUSDG vsr() (chain 4663). The app's (intentionally lossy) BigInt(liqLtv) from the
JSON number is matched exactly, not "fixed".
maxLeverage = calcLeverage(maxLtv); defaultLeverage = calcLeverage(safeLtv),
safeLtv = maxLtv% β 0.75 β exactly as the app derives them. Ladder = integer steps
1x β¦ βmaxLeverageβ (LTV =(1 β 1/lev)Β·100) then the exact max.
- Exit slippage is baked into
collateralTokens.json by the app's weekly refresh script; its file
mtime is the exitLiquidity.asOf. The tier (the only verdict) lives in spiralHints with
its thresholds; raw slippagePct stays under exitLiquidity, null (no route) preserved.