A practical guide to crypto orderbook data (and how to use it with an AI agent)

Most retail traders look at a price chart and call it a day. Professional traders look at the orderbook and learn something very different. The price tells you what has happened; the book tells you what could happen next.

This article is a practical walk-through of crypto orderbooks — what they contain, what questions they answer that nothing else can, and how to query them through an AI agent using the Cryptohopper Market Data MCP.

---

What an orderbook actually is

An orderbook is a list of every resting bid and ask for a market, organised by price level. For each level, you see the price and the total quantity of orders sitting at that price.

Conceptually:

ASKS (sellers waiting to sell)
 Price     Size
 66,180   1.20  ← best ask
 66,200   2.50
 66,220   0.80
 ...
BIDS (buyers waiting to buy)
 Price     Size
 66,175   1.10  ← best bid
 66,160   3.40
 66,140   0.75
 ...

The gap between the best bid and the best ask is the spread. The depth of orders on each side is the depth or liquidity. Everything else — walls, imbalance, slippage — follows from walking this list in one direction or the other.

Orderbooks change constantly. High-liquidity pairs on major exchanges see thousands of updates per second. When you pull one via the Cryptohopper MCP, you're getting a snapshot — a photograph of a flipbook that's mid-motion.

What orderbooks are unbeatable at

Four questions only the orderbook can answer cleanly.

1. Will my order cause slippage? If you want to buy 5 BTC, the best ask might show a size of 1.2 BTC. That takes you 1.2 BTC deep at the best price. The next 1.5 BTC is at a slightly worse price. The remaining 2.3 BTC is further up. Walking those levels gives you your true average fill price, which will be higher than the "last traded price" a ticker reports.

An agent-driven slippage estimator does this walk for you across multiple exchanges and returns a cheapest-venue recommendation. See how to estimate slippage before placing a trade.

2. Is this market actually liquid? Ticker volume can lie. A pair can have high 24h volume because of one large trade five hours ago while the current book is almost empty. The only honest liquidity metric is how much can be transacted right now within X% of mid-price — and that requires the book.

3. Where are the walls? Sometimes one price level holds 10× the size of its neighbours. These walls act as temporary support or resistance. Sometimes they hold; sometimes they break; sometimes they vanish suddenly (spoofing). Either way, traders watch them. See how to detect liquidity walls.

4. What's the order imbalance? If the bid side holds significantly more size than the ask side — or vice versa — that's a short-term sentiment signal. Whether it predicts anything reliably is debatable, but it's information you simply can't get from any other data type.

Five metrics you should know

A handful of derived numbers do most of the heavy lifting when you're reasoning about the book. Each one is a few lines of logic on top of a fresh orderbook snapshot.

Spread — best_ask − best_bid. Usually also expressed in basis points: spread / midpoint × 10_000. Tighter is better.

Depth at X% — cumulative size on the bid (or ask) side within X% of midpoint. Standard comparison is "depth within 2% of mid".

Slippage for order size S — walk the book from the top until you've filled S. The weighted average fill price minus the midpoint, divided by the midpoint, is your slippage estimate.

Imbalance ratio — sum(bid sizes top N levels) / sum(ask sizes top N levels). Values meaningfully above or below 1 signal a skew.

Volume-weighted midpoint — size-weighted midpoint, often used instead of raw midpoint for fair-value comparisons.

When you hand these to an agent you're essentially teaching it to reason about liquidity the way a market-maker does.

Using the orderbook through the MCP

The Cryptohopper MCP exposes an orderbook tool that takes an exchange and a pair and returns a snapshot of the book. Every AI-agent workflow below is one tool call and some local math.

Depth comparison across venues

"Compare the 0.5% depth on both sides for BTC/USDT across
Binance, Coinbase, and Kraken. Tell me which exchange has the
deepest book for buying $100k right now."

The agent pulls three orderbooks, walks each within 0.5% of mid, sums the sizes on the ask side, and ranks them. You get a sentence-long answer in under ten seconds.

Slippage estimation

"I'm about to buy 10 ETH on Binance. What's my expected average
fill price, given the current book?"

The agent pulls the book, walks the asks until it's filled 10 ETH, and tells you the effective price and basis-point slippage. Same call, different interpretation.

Wall detection

"Scan the ETH/USDT book on Kraken. Are there any levels
within 2% of mid-price where the size is more than 5× the
average size of neighbouring levels?"

Agent fetches, computes the rolling mean of level sizes, flags outliers. Would take half a day to build from scratch; three sentences over MCP.

Spread monitor

"Which of the top 20 pairs on Binance has a spread wider than
10 basis points right now?"

Here you're combining ticker data (to find the top 20) with orderbook calls (to measure spread per pair). This is where the MCP's reasoning layer shines — it picks the right tool for each sub-question.

When not to use orderbook data

Orderbooks are heavy. Pulling the full book for 200 pairs every minute is a great way to burn your weekly quota. Rules of thumb:

• Scan broad, act deep. Use tickers to narrow the universe, orderbooks only on the handful you care about.
• Avoid loops in long running agents. A ticker-only loop can fire hundreds of times a day safely; an orderbook loop across many pairs cannot.
• Don't cache and re-use. The orderbook is stale the moment it's fetched. Always re-fetch right before you act.

The rate limits reference has the exact cost profile, and how to feed an agent market context without burning tokens is the companion article for agents that live in a loop.

The two most common mistakes

Mistake 1: Trusting the top of the book. The best bid and best ask are two data points. They are not the whole story of a market. A market can show a tight spread at the top with an empty book underneath, and the moment your size arrives, price moves meaningfully. The only defence is depth analysis — don't stop at level one.

Mistake 2: Ignoring cross-exchange differences. The same pair can trade at noticeably different prices across exchanges, especially altcoins during fast moves. Looking at only one exchange's book hides this. The remedy is to always pull at least two books for any pair you're serious about, and compare them.

What to build on top of orderbooks

• Slippage estimators. The single most useful thing to run before any non-trivial order. Recipe.
• Depth dashboards. One table, one row per exchange, showing 0.5%/1%/2% depth on each side. Great as part of a morning digest.
• Wall detectors. Alert when a new wall appears within a percentage of mid. Recipe.
• Execution advisors. An agent that, given a target order, recommends the best venue and an entry strategy. Combine with the Cryptohopper REST API to let it place the order.

The takeaway

Orderbooks are what separate "I looked at the price" from "I understand the market". They are heavier than tickers and less historically useful than candles, but they are the only data type that answers the single most important pre-trade question: "what will happen if I put a real order in?"

Every serious crypto AI agent should know how to reach for them. The Cryptohopper MCP makes that reach a one-tool call.