There is a difference between asking a chatbot about Bitcoin and plugging a chatbot into a purpose-built data backend. As of today, Backtesting Arena is the latter: a one-click connector inside claude.ai. No API-key fiddling, no copy-pasting of figures. Connect, ask, get numbers from the real pipeline.
This isn't a cosmetic integration. It's an attempt to close a specific gap — the one between "sounds plausible" and "has been computed".
The problem with "Where is the Bitcoin cycle?"
Ask that question of any unconnected language model and you'll get a fluent, plausible answer. It leans on what's out there on the web — articles, forums, repeated rules of thumb. Good enough for a rough orientation. Not good enough for a decision with money at stake.
Because the quantities that actually matter are exactly the ones a web search cannot produce:
- A point-in-time cycle score — the state as it was known on that day, not revised in hindsight.
- A look-ahead-free backtest validation — one that doesn't accidentally bake the future into the past.
- A Deflated Sharpe Ratio — a Sharpe number corrected for the number of variants tested, instead of celebrating the prettiest result out of a hundred tries.
These outputs are not reproducible from a public OHLCV or market-data API. They only exist because of the compute infrastructure behind them. And that is precisely what the connector now exposes.
What's inside: 65 tools
The connector exposes the platform's entire analysis layer. Roughly grouped:
| Category | Examples |
|---|---|
| Cycle & market state | BTC cycle score (10 indicators), Arena Pulse (0–100 heat score), macro regime (18 components) |
| On-chain (since 2009) | MVRV, SOPR, NUPL, Puell, Mayer, whale address cohorts |
| Backtests & edge | validate_strategy (look-ahead-aware, with DSR), Edge-Library filter effects |
| Screening & sentiment | Altcoin screener (Top-200), Fear & Greed, funding rates |
| Interactive (new) | Dip Decision — "buy now or wait for the dip?" |
One bearer token or an OAuth login covers all of it. The core surface is usable on the free tier, no credit card.
The new part: the Dip Decision calculator
Until now the connector pulled data. The new step is interactive tools that compute on your own assumptions. The first one tackles the question everyone knows — buy in now, or wait for the pullback?
The calculator makes no forecast. It takes your assumptions and makes the consequences explicit. An example: you think a target at double is plausible, you'd add on a 50% pullback, you put the probability of that dip at 30% — on €1,000 committed:
| Strategy | Expected value |
|---|---|
| Buy now (all in) | €1,700 |
| Wait for the dip | €1,720 |
| Split 50/50 | €1,710 |
With these assumptions, waiting wins — but barely. And that's the real point: the calculator shows the tipping point. Below a dip probability of roughly 29%, buying now is expected-value-optimal; above it, waiting is. Your 30% sits right on the edge. A second mode additionally computes the risk-adjusted, growth-optimal split (Kelly).
What the calculator does not do: tell you what to do. It gives you the numbers and the tipping point. The interpretation stays with you. That's by design, not a missing feature.
The line we don't cross
Backtesting Arena does not provide investment advice. No tool in the connector says "buy" or "sell". What comes back are distributions, base rates, verdicts with sample size and confidence — and caveats that don't get trimmed away. Under 30 round-trips, every result is treated as anecdote, not evidence, and is labelled as such.
This is deliberate. A tool optimised for attention would hide the caveats. A tool built for good decisions puts them up front. The connector inherits that line unchanged.
How to connect it
In claude.ai:
- Settings → Connectors → Add custom connector.
- Enter the URL:
https://tradingstrategies.work/api/mcp - Follow the OAuth login. The free tier needs no credit card.
For Cursor, Cline or Claude Desktop the same endpoint works via bearer token (free key at /dashboard/account/api-keys). It's a standard MCP server with open OAuth discovery (RFC 9728) — the bearer path works with any MCP-capable client; we've verified the one-click OAuth flow on claude.ai.
After that you can ask directly in a chat — "Where is the Bitcoin cycle, point-in-time?", "Validate this strategy over the last three years", "Buy now or wait for a dip to 0.5×?" — and Claude pulls the answer from the pipeline instead of the web.
What Backtesting Arena contributes here
The value isn't in having "Bitcoin data" — that's everywhere. It's in the compute infrastructure that is hard and boring: point-in-time integrity, overlap and multiple-testing correction, careful cost and out-of-sample handling. That is exactly what can't be rebuilt from a price API — and exactly what now sits behind a connector call. The connector changes nothing about the methodology; it just makes it available where the questions are already being asked.
FAQ
Is this investment advice? No. No tool gives buy or sell signals. The connector delivers distributions, base rates and verdicts with sample size — you make the decision. Backtesting Arena is aimed at informed investors who decide independently.
Does it cost anything? The core surface is usable on the free tier, no credit card. Higher frequency and deeper history sit in the paid tiers (API Pro / API Power) — you don't need them to start.
What data does Claude get about my account? Most tools return public market data and touch no personal data. A few account-specific tools (e.g. your own backtests or subscriptions) are the exception. OAuth access is revocable at any time, and identity data is held in the EU region. Details in the privacy policy.
Why not just ask Claude about Bitcoin without a connector? You can — for a rough orientation it's often enough. The connector shines on the methodically serious question: point-in-time states, DSR-corrected backtests, conditional base rates. That's exactly what a web search categorically cannot deliver.
Does this only work with Claude? No. It's a standard MCP server: the bearer-token path works with any MCP-capable client (Cursor, Cline, Claude Desktop and others). We've verified the one-click OAuth connector on claude.ai; other OAuth-capable clients follow the same open standard (RFC 9728).
What is the Deflated Sharpe Ratio and why do you keep stressing it? The classic Sharpe ratio rewards the prettiest result out of many attempts — the more variants you test, the more likely one looks good by chance. The Deflated Sharpe Ratio (Bailey & López de Prado) corrects for exactly that. It's the difference between "worked in the backtest" and "probably works tomorrow too".
The connector sometimes disconnects — is that known? The OAuth layer is fresh. If you notice a disconnect: reconnect once. We're hardening the token-refresh path continuously.