Prerequisites
Cryptohopper MCP configured in an MCP client — see the setup overview.
Explorer tier or higher. Grid parameter generation requires historical candle data — see subscription tiers.
A Cryptohopper grid-bot configuration if you plan to apply the generated parameters. This guide covers parameter generation, not the bot setup itself.
A rough sense of your intended position size and risk tolerance.
Setup steps
Pick the pair and exchange
Choose a pair you have a view on. Grids work best in range-bound or mildly trending conditions — not during a strong directional move.
Issue the parameter-generation prompt
Using the Cryptohopper MCP, help me configure a grid bot for SOL/USDT on Binance. My position: $5,000 USDT My risk tolerance: moderate (willing to hold through a 20% drawdown but not more) My time horizon: roughly 2 weeks Pull: - The last 200 × 4h candles. - The last 30 × 1d candles. Compute: - The 14-period ATR on the 4h timeframe. - The realised high and low over the last 14 days. - The median daily range over the last 30 days. - The current trend (up, down, ranging). Based on these, propose: - A grid range (upper and lower bound). - Number of grids. - Per-grid size (in USDT). - Whether you would recommend starting this grid right now, or waiting for different conditions. Show your reasoning for each choice.
Review the output
A good response includes a clear range proposal, numeric justification per parameter, and a frank assessment of whether current market structure suits a grid. If the agent spits out numbers without justification, make the prompt more demanding.
Sanity-check against your own intuition
The agent does not know your full portfolio context, P&L history, or events that haven't yet hit candle data. Treat the output as a strong starting proposal, not a final answer.
Apply the parameters to your Cryptohopper grid bot
Use the Cryptohopper dashboard or REST API. The MCP does not execute trades or update bots — see combine MCP + Cryptohopper Trading API for end-to-end agents (NEEDS LINK).
Save the full prompt-and-response log
If the grid performs well, this is documentation of what worked. If it performs poorly, it's a debug trail.
Rules of thumb to encode in the prompt
Bake these (or your own preferred rules) into every run so the agent produces consistent logic rather than reinventing its approach each time.
Parameter | Starting rule |
Range (upper) | Recent 14-day high + ATR(14) |
Range (lower) | Recent 14-day low − ATR(14) |
Number of grids | Such that each grid step ≈ 0.5× ATR(14) on the 4h |
Per-grid size | Position size ÷ number of grids |
Start now? | Only if current trend is "ranging" or the pair is mid-range |
Cost profile
Action | Explorer | Hero |
200 × 4h candles | 5 units | 1 unit |
30 × 1d candles | 5 units | 1 unit |
Per configuration | ~10 units | ~2 units |
Running daily for a small set of pairs costs under 100 call units/week. See rate limits explained.
Troubleshooting
The proposed range is wildly too wide or too narrow
The ATR multiplier or lookback is wrong for the current regime. In a high-volatility week, 200 × 4h captures an unusual range that won't hold; in a quiet week, it's too tight. Try restricting to the last 60 × 4h candles, or instruct the agent to prefer the narrower of (recent high/low) and (volatility-based range).
The agent proposes a grid during a strong trend
Grids fail in trending markets. Add to the prompt: "If the current trend is strongly up or down (e.g. >5% in the last 48 hours), recommend against starting a grid and suggest waiting for consolidation."
Per-grid size calculations look off
Have the agent show the arithmetic explicitly: "For each grid, state: grid N of M, price range from X to Y, size in USDT, size in base asset, expected fills if the bot rotates through the full range once."
HISTORY_LIMIT_EXCEEDED
The lookback exceeds your tier. Reduce to 200 × 4h or 60 × 1d as a baseline. See subscription tiers and candle tool reference.
The output ignores your risk tolerance or time horizon
These are soft inputs — the model may not weight them heavily. Re-emphasise at the end of the prompt: "Before proposing a range, explicitly state how your proposal matches my 20% max drawdown tolerance and 2-week horizon."
The recommendation changes significantly between runs
Normal — candle data evolves. For a stable recommendation for a paper-bot test, cache the first run's output. For daily regime-aware updates, re-running is the correct behaviour.
You want this fully automated, including applying to the grid
The MCP generates parameters; the Cryptohopper REST API applies them. See research via MCP, execute via REST API (NEEDS LINK).