AI calc infrastructure

Don't trust AI-written backtest.
Use verified primitives.

AI に Sharpe を何度も再実装させない。

Deterministic finance primitives for Claude, Cursor, Codex, ChatGPT — backed by 9 months of dogfooded FX backtest engine.

AI に計算させるのではなく、 AI が呼ぶ計算 infra を提供。 MCP server + REST API 両経路、 7 日 trial → 月 ¥980。

Get License Key View API docs

Same numbers every time. 100× faster than AI rewriting. Audit-ready.

claude-desktop

You: Sharpe ratio of EURJPY with RSI(14) strategy?

Claude: calling fx_backtest.run

{
  "strategy": "rsi_oversold",
  "symbol":   "EURJPY",
  "params":   { "rsi_period": 14 }
}

✓ verified, deterministic, 187 ms

Claude:

Sharpe:      1.42
Max DD:      -8.3%
Win rate:    54.1%
Profit factor: 1.78

How it works

Stop letting AI reinvent your Sharpe ratio.
3 steps to deterministic finance.

I built this because I was tired of Claude reinventing Sharpe every chat session. Now my AI agent calls verified primitives instead.

Get a license key
Sign up on the dashboard, choose a plan, and get an API key + MCP license in under 60 seconds.
ダッシュボードで sign up → Stripe で plan 選択 → API key + MCP license を 60 秒で取得。
Wire it into your AI
Add the MCP server to Claude Desktop / Cursor / Codex config, or use the REST API directly from any framework.
Claude Desktop / Cursor / Codex の config に MCP 追加、 or 任意言語から REST API を直接 call。
Ask. AI does the rest.
Tell your agent what to test. It fetches data, calls our verified primitives, and returns deterministic results.
AI に検証内容を質問するだけ。 data 取得 + 計算 + 結果整理を AI が orchestrate。

Calc primitives

8 primitives. Same numbers every time.

Verified engine, not AI-generated code. 9 months of dogfooded development. 303 unit tests + multi-round adversarial review.

9 ヶ月投資の検証済 engine。 303 件 test + Codex 多 round review。

run_backtest
Execute strategy across OHLC, return trades + equity curve.
戦略を OHLC で実行 + trade + equity curve 返却。
calc_sharpe
Risk-adjusted return — annualized, deterministic, audit-ready.
risk-adjusted return = 年率化 + 決定的 + 監査可能。
calc_maxdd
Peak-to-trough drawdown + duration + recovery time.
最大 drawdown + 期間 + 回復時間。
run_walkforward
Anchored or rolling WFO with purged splits — no lookahead.
anchored / rolling WFO + purged split = look-ahead なし。
run_montecarlo
Bootstrap returns, get distribution of outcomes + percentiles.
bootstrap で結果分布 + percentile 返却。
optimize_grid
Brute-force parameter sweep with explicit budget cap.
パラメータ grid 全探索 = 明示上限付き。
optimize_genetic
Evolutionary search over high-dim parameter space.
遺伝アルゴリズムで高次元パラメータ探索。
compare_strategies
Side-by-side stats across N strategies, single call.
N 戦略を 1 call で並列比較 = 統計 / metric matrix 返却。

Two transports, one engine

Reach the same engine from anywhere your AI lives.

Call our primitives from Claude Desktop, Cursor, Codex, ChatGPT Custom GPTs, your own LangChain agent — or plain curl. Same shared core, two transports.

同計算 engine = 2 transport で提供。 AI agent でも任意言語でも同 result。

MCP server

stdio protocol

Plug into Claude Desktop, Cursor, Codex CLI, ChatGPT Desktop, or any MCP-compatible IDE.

Claude Desktop / Cursor / Codex CLI / ChatGPT Desktop / 任意 MCP 対応 IDE に config 1 行で統合。

{
  "mcpServers": {
    "quantprims": {
      "command": "npx",
      "args": ["@quantprims/mcp"],
      "env": { "LICENSE_KEY": "fxb-…" }
    }
  }
}

REST API

HTTPS + JSON

Call from any framework, language, or web UI. Same primitives, generic HTTP transport.

任意言語 / framework / web UI から HTTP で call。同 primitives、汎用 transport。

curl https://api.quantprims.com/v1/calc_sharpe \
  -H "Authorization: Bearer fxb-…" \
  -d '{ "returns": [...] }'

Both backed by the same shared/ core — 9 months of dogfooded TypeScript primitives.

Why trust this

AI is fast. AI is wrong sometimes.
Wolfram-grade primitives fix that.

Same story as Wolfram Alpha for math — when correctness matters, you call a verified engine, not let the LLM reinvent it.

Wolfram Alpha が GPT 時代でも数式 path で棲み分けを保つ同構造。

Deterministic
Same input → same output, every time. No AI hallucination drift between sessions.
同 input = 同 output 保証。 AI 毎回別実装で数値ブレる path 撤廃。
100× faster iteration
Verified primitives return in ~0.1s. AI re-implementing Sharpe each time takes 5-20s.
検証済 primitives = 0.1 秒。 AI 毎回 code 書く path 比 100×。
Audit-ready
9 months of TypeScript primitives. 303 unit tests. Multi-round adversarial review log.
TypeScript primitives 9 ヶ月投資 + 303 件 test + Codex 多 round review 履歴。
Safe boundary
Schema-constrained inputs + read-only data. No prompt injection, no arbitrary execution.
schema 制約 input + read-only data = prompt injection 隔離、任意実行 risk なし。

Pricing

Founder dogfood pricing.

This is what I pay myself. One plan, all features. Try 7 days free.

founder 本人が毎日払う想定 price。 1 plan + 全機能同 access、 7 日 trial 後課金。

Free trial

7 days, no card required

¥0for 7 days

All 8 calc primitives
MCP server + REST API
Up to 2 devices per license
Audit log + reproducibility
Email support

Start trial

Recommended

Pro

Founder dogfood pricing

¥980/ month

All 8 calc primitives
MCP server + REST API
Up to 2 devices per license
Audit log + reproducibility
Email support

Get license key

※ 商用利用について = 投資助言業非該当確認 (= 弁護士書面取得 path) を進行中。数字だけ返す path (= 評価 channel 削除) で設計済。利用規約参照。

FAQ

Common questions

Can I use this from ChatGPT, not just Claude?ChatGPT でも使えますか？ Claude 限定では？

Yes. The REST API works from any HTTP client — Claude, ChatGPT Custom GPTs (Actions), Gemini, Cursor bash, LangChain, plain curl, your own SDK. MCP transport additionally covers Claude Desktop / Cursor / Codex CLI / IDE clients.

YES = REST API は任意 HTTP client (= ChatGPT Custom GPT Actions / Gemini / curl / LangChain / 任意言語 SDK) から call 可能。 MCP transport は Claude Desktop / Cursor / Codex CLI / IDE client carry。

How do I wire it into Claude Desktop?Claude Desktop に設定する方法は？

Add a single block to claude_desktop_config.json with the MCP server command + your LICENSE_KEY env var. Full setup snippet on the docs page. Takes under 2 minutes.

claude_desktop_config.json に MCP server 起動 command + LICENSE_KEY env を追加するだけ。詳細 sample = docs ページ参照、 2 分以内完了。

Can I use this commercially?商用利用可能ですか？

Single-license commercial use (= personal trading research, consulting prep) is intended. We are completing a lawyer-signed opinion on non-applicability to Japanese investment advisory law (= 投資助言業) before formal launch. Until that is published, treat commercial use as best-effort.

個人商用利用 (= 自己投資検証、 consulting 準備) 想定。日本法投資助言業非該当弁護士書面取得 path 進行中、 launch 前公開予定。それまで商用利用は best-effort。

Isn't this giving investment advice?これは投資助言では？

No. The engine returns numbers (= Sharpe / max drawdown / win rate / etc.) — never recommendations, evaluations, or buy/sell signals. The 'number-only' path was a 13-round refactor specifically to drop investment advisory law risk from 20-35% to 2-5%. Interpretation is on you (or your AI agent).

NO = 数字 (= Sharpe / max DD / win rate / etc.) のみ返却、推奨 / 評価 / 売買 signal なし。「数字だけ返す」 path = 13 round 改修で投資助言業該当 risk 20-35% → 2-5%。解釈は user (or AI agent) 責任。

Where does the price data come from?価格 data の出所はどこですか？

We don't supply price data. You (or your AI agent) pass OHLC arrays as input — sourced from your broker export, paid data API, or your own collection. Same structure as Wolfram Alpha for math: bring your numbers, we calculate. Data source compliance is your responsibility.

data 提供しません。 OHLC array を input で渡す path (= broker export / 有料 API / 自前収集から)。 Wolfram Alpha 同構造 = 数式を渡すと計算結果を返す。 data 出所規約適合性は user 責任。

What's covered by the license?license でカバーされる範囲は？

All 8 calc primitives, both transports (MCP + REST), up to 2 devices per license, full audit log + reproducibility. No feature lock between trial and paid — only the time window changes.

8 primitives 全 + MCP + REST 両 transport + 2 device / license + audit log + reproducibility。 trial と paid 間で feature lock なし、期間制限のみ。

What happens to fx-backtest.com web simulator?既 fx-backtest.com web simulator は？

Legacy web simulator remains available at /simulator for existing users. New development focus is on the MCP + REST API path. Web UI gets maintenance-only.

既 user 向けに /simulator 維持 (= freeze)。新開発 = MCP + REST API path 集中、 web UI は maintenance only。

Don't trust AI-written backtest.Use verified primitives.

Stop letting AI reinvent your Sharpe ratio.3 steps to deterministic finance.

Get a license key

Wire it into your AI

Ask. AI does the rest.

8 primitives. Same numbers every time.

Reach the same engine from anywhere your AI lives.

MCP server

REST API

AI is fast. AI is wrong sometimes.Wolfram-grade primitives fix that.

Deterministic

100× faster iteration

Audit-ready

Safe boundary