MCP connector
What you see in chat
Spera results arrive as a live card that answers the question you asked — one chart or graph you can point at, the evidence drawn where it happened, how far a result has been tested, and what apps that can't show cards display instead.
Updated 2026-10-16·Rev. 2026.06
When your assistant builds, opens, or tests something in Spera, the result comes back as a card: a small Spera panel inside the conversation. The assistant adds a sentence or two of interpretation; the card holds the figures, so you are not reading a block of raw data.
Every card answers one question
| Card | The question | What it shows |
|---|---|---|
| Strategy | What did you build? | The strategy read left to right: what it reads, when it acts, and what it does |
| Result | Does it work? | The equity curve against buy and hold, with return, max drawdown, and trades above it |
| Change | Did my change help? | The new run and the previous one on the same axis, and how far apart they are |
| Compare | Which branch should I keep? | Every branch on one chart, with standings beside it |
While a deep backtest is running, the same card shows the engine's own stages; once the run is over, its walk-forward periods appear while the result is saved. If something needs your input, the card opens on the part that needs it. If something fails, it says what happened in one plain sentence.
One chart, three figures
A result card is read through one chart. Above it, three figures follow wherever you point:
- Point or scrub (or use the arrow keys) to read any date. The figures show both lines on that day and how far apart they were.
- Drag across the chart to select a period. The figures re-read for that stretch alone — its return, its worst drop, its length — and Clear returns to the whole run. On a finished deep backtest, List the trades in this period shows the trades entered in it: how many, how many won, the best and the worst, and the first six with their side, length, and return. They come from the same run, read by the card itself, so the assistant is not asked. Apps that do not let a card read results do not show the button.
- Lenses switch what the line shows: Return, Drawdown (how far below its high it was), and, on longer runs, Months (a calendar of monthly returns).
The deepest drop is pinned on the line with its figure, so the largest risk is visible without looking for it.
Evidence, drawn where it happened
Below a deep backtest's chart, each check that ran is a chip. Picking one draws it on the chart and says it in one line:
- Holdout — shades the stretch the strategy never saw and compares how it did there.
- Walk-forward — marks each period tested on its own, with its return.
- Market — shades the rising, ranging, and falling stretches of the window, and says how the trades entered in each did.
- Edge Check — brackets where randomly timed copies of the strategy ended, beside where this run ended. If it has not run yet, the chip offers to run it.
Strategy cards
A strategy, module, or custom node is drawn as its graph. Pick a node to light what feeds it and what it drives; a sentence under the graph says what it does. After an edit, the card opens on what that edit changed — new nodes tagged, changed values shown as was → now, removed nodes listed — with the whole strategy one tap away. When something is missing, the card opens on the node that needs it.
The trust ladder
Three rungs, filled from the left:
- Not tested yet — it compiled, and that is all anyone knows.
- Past data only — a backtest ran, but only on the data the strategy was built against. Quick backtests always stay here, because they cannot be anything else.
- Held up on unseen data — a deep backtest's holdout period held up, the walk-forward periods did not collapse into one lucky stretch, and the track record is long enough to read.
Only the third rung is shown in colour. Under the rung, each check that ran is listed with its reading, so when something holds a result back — the holdout did not hold up, most of the return came from a few periods, or the track record is too short — you can see which.
The Edge Check's reading is listed with the others, but it does not move the rung. See What to ask for.
A card never grades your strategy
The wording states what was measured and which way it moved — "return is 1.2 points lower", never "worse". Deciding whether that trade-off is worth it is a conversation, not a label.
Where the numbers come from
Under the trust rung, a result names its source: the Spera engine, how many candles it ran on, the revision it tested, and the run, with a link to the strategy in Spera. Every figure on the card comes from that run. None is written by the assistant. The limits every card is held to — its size, what it costs your assistant to read, its height — are in What every card is held to.
What to ask next
Each card offers one suggested next step, and beside it up to two smaller questions, each labelled with what it is about: the month of the deepest drop, the weakest walk-forward period, the market phase that paid most, or how a strategy enters and exits. When you select a period or pick an evidence chip, the main button becomes a question about what you selected, such as "Ask about Mar 4 – Apr 18". Selecting any of them sends that question into the chat, as if you had typed it.
Comparing a change
When a deep backtest finishes, Spera looks for the previous completed deep backtest of the same strategy branch, pair, timeframe, market, exchange, and date range, run with the same backtest settings (starting capital, fees, slippage, position sizing, and warmup) and made from a different revision. If there is one, you get a Change card instead of a Result card, and its walk-forward chip shows how much the change helped in each period.
Anything less alike — another date range, another pair, different fees, the same revision run twice — would put two unlike numbers side by side and credit the difference to your change, so there is simply no comparison. Quick backtests are never compared.
Which apps show the card
Claude, ChatGPT, Cursor, and VS Code display the card itself, in light or dark to match the app. Results from before these live cards keep showing the earlier, simpler card.
Apps that do not display cards — Codex and Claude Code in the terminal, for example — show the same card as a short block of text: the answer, a small line chart made of characters, the figures and evidence as lines, the trust rung as dots, and the suggested next step. The full result follows it, so nothing is lost.
What a card cannot do
A card only shows, reads, and suggests. The one thing it reads on its own is a selected period's trades, from the run it is showing. It cannot start a backtest, change a strategy, or trade — every change still goes through the conversation, with the same permissions you approved when you connected. "Open in Spera" takes you to the strategy in the app.
Next
- Three things worth doing first
- What to ask for
- Backtesting — what the figures mean