Backtesting with your agent
You don't need a data pipeline to test a strategy idea. Connect DataMaxi+ to your agent, describe the strategy, and the agent pulls the history it needs, writes the test and shows you the result. You stay in charge of the assumptions; the agent does the plumbing.
What it costs
History calls are cheap and capped per call, so you can estimate the bill before you start:
| Data | Per call | 1 year × 5 venues | Cost |
|---|---|---|---|
| Funding history | 100 rows | 11 calls per venue: 55 | $0.0275 |
| Daily candles | 1,000 candles | 1 call per venue: 5 | $0.0025 |
| Total | ≈ 60 History calls | ≈ $0.03 |
A year of 8-hour funding settlements is 1,095 rows per venue, so 11 pages of 100. Each History call (funding_rate_history, cex_candle) costs $0.0005.
The $5 sign-up credit covers about 160 backtests of that size. Hourly settlements or 1-hour candles take more pages: 8,760 hourly funding rows per venue is 88 calls, and a year of 1-hour candles is 9 calls of 1,000.
Limits to plan around (also on Rows per call):
- Funding history: at most 100 rows per call; page with
pageor movefrom/to. - Candles: at most 1,000 per call, and only from 2024-01-01.
- Liquidation history per symbol:
15mbuckets cover up to 72 h,1hbuckets up to 7 days, so liquidations suit short event studies, not multi-year tests.
1. Connect DataMaxi+
Add the MCP server to your client with your API key (one command for Claude Code; full setup for Claude, ChatGPT and Cursor on the MCP page):
claude mcp add datamaxi --transport http https://mcp.datamaxiplus.com/mcp \
--header "X-DTMX-APIKEY: YOUR_API_KEY"
2. Describe the test
Be specific about the universe, the window, the rules and the costs. For example:
Backtest a delta-neutral funding carry on BTC-USDT perps over the last 12 months on Binance, Bybit, OKX, Bitget and Gate. Pull the funding history with DataMaxi+ (
funding_rate_history, 100 rows per call, page until you have the full year) and daily candles withcex_candle. Each day, hold the venue with the highest trailing 7-day average funding; switch only when another venue beats it by 5% APR. Charge 0.1% per switch. Report annualized return, max drawdown and number of switches, and plot equity versus always holding Binance. Tell me how many DataMaxi+ calls you made.
A good agent will:
- Call
funding_rate_exchanges/funding_rate_symbols(free) to confirm the venues and symbol. - Page
funding_rate_historyper venue until the window is covered, and fetch candles withcex_candle. - Run the test in its own sandbox or your local Python, and show the code so you can check it.
- Optionally call
get_usage(free) to report what the run cost.
3. Check the result like a skeptic
- Look-ahead. Signals must only use data available at decision time (trailing averages, not centered ones).
- Costs. Include fees, slippage and transfer time. Funding carry that survives 0.1% per switch is more believable than one that needs zero.
- Settlement intervals. Venues settle every 1, 4 or 8 hours; compare in APR, not per-period rates.
- Coverage. A venue that listed the pair mid-window has a shorter history; make the agent say so.
Doing it yourself over the API
The same pull in plain Python, if you'd rather run the test in your own code:
- Python
import os
import time
import requests
API = "https://api.datamaxiplus.com/api/v1"
HEADERS = {"X-DTMX-APIKEY": os.environ["DATAMAXI_API_KEY"]}
VENUES = ["binance", "bybit", "okx", "bitget", "gateio"]
FROM = "2025-11-20"
TO = "2026-11-20"
def funding_history(exchange, symbol="BTC-USDT"):
rows, page = [], 1
while True:
r = requests.get(
f"{API}/funding-rate/history",
headers=HEADERS,
params={"exchange": exchange, "symbol": symbol, "from": FROM, "to": TO,
"limit": 100, "page": page, "sort": "asc"},
timeout=15,
)
if r.status_code == 402:
raise SystemExit("Balance empty: top up at https://datamaxiplus.com/dashboard/billing")
if r.status_code == 429:
time.sleep(int(r.headers.get("Retry-After", "60")))
continue
r.raise_for_status()
batch = r.json()["data"] # [{"d": unix_seconds, "f": rate}, ...]
rows += batch
if len(batch) < 100:
return rows
page += 1
history = {v: funding_history(v) for v in VENUES}
for venue, rows in history.items():
apr = sum(r["f"] for r in rows) * 100 # sum of one year of settlements, in %
print(f"{venue:8s} {len(rows):5d} rows {apr:6.1f}% simple annual funding")
Each page is one History call ($0.0005), whatever it returns. 429 responses are free; 402 means the balance is empty.
Related
- Funding rate arbitrage — the strategy behind the example.
- Pricing and billing — tiers, packs and what happens at zero.
- MCP server — every tool, including
get_usage,topupandcontact_datamaxi. - Missing data for your test? Ask your agent to request it with
contact_datamaxi.