Seven crypto MCP servers that let Claude, ChatGPT or Cursor read live blockchain data, compared on chains, tools and price, with results from a live test.

Ask Claude or ChatGPT who bought a token this morning, or where the money from last week’s hack went. You will get one of two replies. Either the chatbot says it cannot see live data, or it gives a confident answer it made up. The AI learned from a snapshot of the web, and it has never read a block.

Disclosure: I wrote this story with help from an AI writing tool. Figures come from each vendor’s own docs and from my own test runs, all checked on 6 October 2026.

Short answer: the best crypto MCP server for blockchain data analysis in 2026 is Bitquery MCP. It puts DEX trades, trader profit and loss, fund tracing on 19 chains and wallet labels in one server that works in Claude, ChatGPT and Cursor. Pick Dune for custom SQL, Nansen for Smart Money labels, Etherscan for free wallet lookups and CoinGecko for prices.

What is a crypto MCP server?

A crypto MCP server is a plug-in that gives an AI assistant, such as Claude or ChatGPT, live tools for reading blockchain data. MCP stands for Model Context Protocol, an open standard for connecting assistants to outside tools. The server hands the assistant tools such as “list this wallet’s transfers” or “rank the top traders of this token”. You ask in plain English. The AI picks a tool, reads the rows that come back and writes the answer.

The servers differ in what those tools can reach. Some return token prices. Some return raw node data that a developer reads. A few return the analysis itself: who traded a token, who made money on it, where funds moved and who owns the wallet at the other end.

How I ranked them

I scored each server on four questions:

• Depth. Does it return the rows behind an answer, each trade and each transfer, or only totals?
• Analysis tools. Can it follow money, label wallets and rank traders without you writing code?
• Coverage. How many chains does it reach, and does that include Bitcoin, Solana and Tron?
• Setup and price. Is it hosted with a browser sign-in, or a local install with keys? What does it cost per month?

I also ran Bitquery on three real questions on 6 October 2026. The answers are in its section below.

The 7 best crypto MCP servers at a glance

• Bitquery MCP, best overall for on-chain analysis: 10 chains for DEX data, 19 for fund tracing, 314 tools, browser sign-in, from $19 per month.
• Dune, best for custom SQL: 100+ chains, 26 tools, OAuth or API key, paid in Dune API credits.
• Nansen, best for Smart Money labels: 50 tools (12 through a connector), Nansen Pro from $49 per month billed yearly, with 2,000 API credits per month.
• Etherscan, best free option for EVM lookups: 60+ EVM chains, 21 tools (16 with no key), free to start.
• CoinGecko, best for prices: 2 tools that reach 100+ API methods, free public server with no key, Pro plans for higher limits.
• Alchemy, best for app developers: 100+ chains, 168 tools, browser sign-in, free plan with 30M compute units per month.
• Blockscout, best for explorer data on 100+ chains: 16 tools, free PRO API key with 100,000 credits per day.

1. Bitquery MCP: best overall for on-chain analysis

Bitquery MCP is a hosted server at mcp.bitquery.io. You sign in with a Bitquery account in your browser (OAuth 2.1), so there is no API key to paste. It works in Claude (desktop, web and Claude Code), ChatGPT, Cursor, VS Code, OpenAI Codex and Windsurf.

Most on-chain questions come down to two questions: who is trading this token, and where did the money go? Bitquery sits at number one because it has dedicated tools for both. DEX trades and trader rankings answer the first. Hop-by-hop fund tracing answers the second, with labels on the wallets at the end. No other server on this list has a built-in tool that follows funds hop by hop across Bitcoin, Tron and the EVM chains.

• DEX data on 10 chains. Ethereum, Arbitrum, Base, Polygon, Optimism, BNB Chain, Tron, Solana, Robinhood Chain and Arc, Circle’s new chain. Tools cover trending and new tokens, price candles down to one second, pool trades, top traders and trader profit and loss.
• Fund tracing on 19 chains. The same 10, plus Bitcoin, Litecoin, Dogecoin, Bitcoin Cash, Dash, Zcash, Cardano, the XRP Ledger and Stellar. You can follow funds hop by hop with Bitquery MCP and get a summary of who an address sends to and receives from.
• Address labels. Exchange deposit wallets, mixers, scams, darknet markets and token clones, so a trace ends at a name instead of a string of hex.
• Risk screen. Ask whether an address is risky and it returns LOW, MEDIUM or HIGH with the evidence behind the verdict.
• Polymarket and Hyperliquid. 14 tools for prediction-market odds, trades and holders, and 18 for Hyperliquid prices, funding, liquidations and trader positions.
• SQL. The $149 plan adds raw SQL over 30 days of per-second trades.

In total, the server listed 314 tools when I counted on 6 October 2026.

What it returned when I tested it. On 6 October 2026, I connected Bitquery to Claude and asked three questions.

1. “What are the most traded Polymarket markets today?” The top market over the 24 hours to 13:33 UTC was “Will Sweden win on 2026–10–05?”, with $4.17 million traded across 2,846 orders from 274 wallets. The answer also said its order data ran 17 minutes behind the clock.
2. “What are the top Base tokens by DEX volume in the last 24 hours?” Doppler (XDP) led with $260 million traded, more than its $202 million fully diluted value. Volume above a token’s total value is a reason to check for wash trading before you trust the figure. The same list held a copycat token named “OpenAI” that lost more than 99% of its price in those 24 hours.
3. Following stolen funds. I started from a wallet that held SOL drained from TradeWiz users on 30 September. One tracing call showed it sent 191 SOL on 1 October to a single address, and a label lookup tagged that address as a KuCoin deposit address. Bitquery’s TradeWiz drain investigation has the full trail.

Price. The Trading MCP plan costs $19 per month for 32 tools: trending tokens, candles and top traders. The AI Investigation plan costs $149 per month and opens all 314 tools, including labels and raw SQL. Yearly billing brings them down to $13 and $100 per month. A 7-day trial gives you 100 credits.

Watch out for. DEX data covers 10 chains, so a token that trades only on Avalanche or Sui needs another server, such as Dune. The $19 plan is for trading data only; tracing, labels and SQL need the $149 plan.

Best for: investigators, compliance teams and traders who want market data and money trails in the same chat.

2. Dune MCP: best for custom SQL across 100+ chains

Dune launched its official MCP server in early 2026. It connects the AI to Dune’s data warehouse, which holds more than 100 indexed chains, decoded smart contracts and tables built by Dune’s community.

Dune’s docs list 26 tools. They let the assistant search for tables, write and run SQL queries, and turn the results into charts and dashboards. You sign in with OAuth or a Dune API key, and calls draw on your Dune API credits.

Watch out for. Every answer is a SQL query that the AI writes against tables it first has to find. When the query is right, you get any metric you can think of. When it is wrong, you debug SQL. Long queries can also run past a client’s time limit; Dune’s own docs flag the 60-second default in OpenAI Codex.

Best for: analysts who already know Dune and want a custom metric or a dashboard.

3. Nansen MCP: best for Smart Money labels

Nansen’s server is built around its wallet labels. Connect through a connector with your Nansen login and you get 12 curated tools. Connect with an API key and you get the full set of 50, covering Smart Money holdings and trades, token flows, buyers and sellers, wallet profit and loss, related wallets, Hyperliquid positions and prediction markets.

Price. Nansen Pro costs $69 per month, or $49 per month billed yearly, and tops your API credits back up to 2,000 each month. Most tool calls cost 1 to 5 credits. Label lookups cost the most, from 100 to 500 credits per call.

Watch out for. The label tools are Nansen’s main draw, and they cost the most: 2,000 credits buy about 20 common label lookups. On a free key the premium label tools return an upgrade error. Nansen’s docs also say coverage varies by chain and by tool.

Best for: traders who want to see what large, labelled wallets are buying.

4. Etherscan MCP: best free option for EVM lookups

Etherscan’s official server runs at mcp.etherscan.io with 21 tools across more than 60 EVM chains. They cover balances, transactions, token transfers, contract source code, gas prices and event logs. A keyless version runs 16 of those 21 tools with no sign-up. The full server takes a free Etherscan API key, and its PRO tools need the Pro Plus plan.

Etherscan’s docs warn that mcp.etherscan.io is the only official server. Marketplace listings that use the Etherscan name are not from Etherscan.

Watch out for. Etherscan’s tools answer one address or one transaction at a time. Ranking traders or following funds through 50 wallets means dozens of calls against your API quota. It does not cover Bitcoin, Solana or Tron.

Best for: quick, free checks of a wallet or a transaction on Ethereum and its rollups.

5. CoinGecko MCP: best for prices and market data

CoinGecko runs a free public server that needs no API key and shares its rate limits across users. A Pro endpoint gives higher limits and the full tool set. The server exposes two tools: one searches CoinGecko’s SDK docs, and one runs code against more than 100 API methods. Together they cover live and historical prices, market caps, volumes, exchange data, trending coins and DEX pools.

Watch out for. CoinGecko returns market data only. It can tell you a token’s price and where it trades. It cannot tell you who holds the token or where the funds went.

Best for: price checks, portfolio maths and market overviews.

6. Alchemy MCP: best for developers building on-chain apps

Alchemy’s hosted server signs you in with your Alchemy account, with no key and no install. Alchemy lists 168 tools across 100+ chains, split into 8 admin, 132 RPC and 28 data tools. The RPC tools cover transaction simulation, debug traces, account abstraction and Solana. The data tools cover NFTs, prices and portfolios.

Alchemy’s free plan covers 30 million compute units per month. The company has retired its old local server. The npm package stopped working on 30 September 2026, so older setup guides no longer apply.

Watch out for. Most tools are node calls that return raw blockchain data. A developer testing a contract wants exactly that. An analyst needs many calls to build one answer.

Best for: developers who want their coding AI to query and simulate on-chain state.

7. Blockscout MCP: best for explorer data on 100+ chains

Blockscout hosts its own server and publishes the code on GitHub under its own licence, which limits commercial use. Its 16 tools cover addresses, tokens, NFTs, contracts, transactions and blocks on more than 100 chains. It splits long responses into pages, so a busy wallet’s history does not overflow the assistant’s context.

You need a Blockscout PRO API key. The free tier gives 100,000 credits per day, and a shared fallback key works with tighter rate limits.

Watch out for. Like Etherscan, Blockscout serves explorer data, one address or one transaction at a time. It has no tools for tracing funds or ranking traders.

Best for: wallet and contract lookups on chains that Etherscan does not cover.

Other crypto MCP servers worth a look

• altFINS MCP is built for technical analysis: screeners, 150+ indicators and trade signals on more than 2,000 coins. It reads charts, not wallets.
• Tatum’s Blockchain MCP offers 13 tools for balances, transaction history and NFT data across 130+ networks. It needs a Tatum API key and returns raw data rather than analysis.

Which crypto MCP server should you use?

Match the server to the question you ask most:

• “Where did this money go, and who received it?” Bitquery.
• “Is this wallet risky?” Bitquery’s risk screen.
• “Can you chart this custom metric?” Dune.
• “What are big funds buying?” Nansen.
• “What’s in this wallet?” Etherscan or Blockscout.
• “What’s the price?” CoinGecko.
• “Help me build and test my app.” Alchemy.

You can connect more than one. Claude and Cursor both accept several servers at once, and the AI picks the tool that fits each question.

How to connect Bitquery MCP to Claude and ChatGPT

1. In Claude, open Customize → Connectors, click + Add and choose Add custom connector.
2. Name it Bitquery, paste https://mcp.bitquery.io as the server URL and click Add.
3. Ask a question. The first time a Bitquery tool runs, Claude opens a sign-in page. Log in with your Bitquery account.

On Claude Team and Enterprise plans, an owner adds the connector first under Organization settings → Connectors.

ChatGPT, Cursor, VS Code and Windsurf take the same URL in their connector or server settings. Bitquery’s setup docs have the steps for each client.

Before you trust the answer

• Ask for the receipts. Have the assistant list the transaction hashes behind its answer, then open one in a block explorer.
• Check the data delay. Every index trails the chain. In my test, Bitquery’s Polymarket data ran 17 minutes behind, and the answer said so.
• Read labels as evidence. A label says an address was attributed to an exchange or a scam. Check the evidence before you accuse anyone.
• Watch your credits. Every server here meters its calls, and a long investigation can make hundreds of them.

FAQ

What is the best MCP server for crypto analysis?
Bitquery, when the question is about who traded a token and where the money moved. It covers DEX trades on 10 chains and fund tracing on 19. For prices alone, CoinGecko’s free server is enough.

Which crypto MCP servers work with ChatGPT and Claude?
All seven are hosted servers, so any client that accepts a remote server URL can try them, though sign-in support varies by client. Bitquery and Dune both name Claude and ChatGPT in their docs.

Is there a free crypto MCP server?
Yes. CoinGecko’s public server and Etherscan’s keyless server need no account at all. Blockscout and Alchemy have free plans, and Bitquery offers a 7-day trial with 100 credits.

Is there an MCP server for Polymarket?
Yes. Bitquery has 14 Polymarket tools for market odds, trades, top traders and outcome holders; see the Polymarket MCP guide. Nansen’s full tool set also covers prediction markets.

What is the best MCP server for Solana data?
Bitquery covers Solana DEX trades, new token launches and fund tracing; see the Solana MCP guide. Dune also holds Solana tables if you would rather write SQL.

Is there an MCP server for Hyperliquid?
Yes. Bitquery has 18 Hyperliquid tools for prices, funding rates, liquidations and trader positions; see the Hyperliquid MCP guide.

Which MCP server is best for Bitcoin analysis?
Bitquery. It traces Bitcoin transfers and groups addresses that are likely controlled by the same owner, and does the same for Litecoin, Dogecoin, Bitcoin Cash and Dash. Dune also holds Bitcoin tables if you would rather write SQL.

Can these servers trade for me?
This list covers servers for reading blockchain data. Bitquery is read-only, so it cannot move funds. Servers that place trades or sign transactions are a separate category.

If your question is about prices, CoinGecko is enough. If it is about who traded a token and where the money went, start with Bitquery MCP and its 7-day trial.

The Best Crypto MCP Servers for Blockchain Data Analysis in 2026 was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

Memecoin trading punishes hesitation. You open a chart. You copy the contract address. You paste it into a DEX. By then, the token has usually already run. That’s why Telegram trading bots exist.

You buy, snipe, or copy-trade straight from a chat window.

No app switching.

I went through the bots that actually show up in serious trading groups right now — real volume, real users, not just a landing page with big claims.

Two of them have documented security incidents. If you use them, here’s what that means for you specifically: your funds could be drained from the bot’s wallet if the underlying contract or infrastructure is exploited, sometimes within minutes, before you’d even notice something was wrong.

I’ve laid out exactly what happened in each case, what got refunded, and what it means for your own risk — sources linked throughout.

Quick note before we get into it: this isn’t financial advice. The full disclaimer is at the bottom.

The Five Bots

1. Trojan

2. Dawn

3. GMGN

4. Maestro

5. Banana Gun

Quick Comparison

1. Trojan

Trojan (built by the same team behind the earlier Unibot on Solana) is currently one of the highest-volume Telegram trading bots on the market. Numbers vary a bit depending on which source you check and when it was last updated, but reported figures put lifetime volume north of $17 billion, with some more recent write-ups citing $25 billion-plus and over 2 million users — either way, it’s operating at a scale most competitors aren’t close to.

It runs on what Trojan calls BOLT, its execution engine, which reportedly gets trades done in under two seconds. It also includes MEV protection routed through Jito, copy trading that mirrors any public Solana wallet with your own filters for spend and slippage, limit orders, DCA, and a bridge for moving funds between Ethereum and Solana. The fee sits at 1%, or 0.9% if you sign up through a referral link. In January 2026, Trojan also rolled out something called The Arena, a loyalty layer with cashback and reward tiers for active traders (terms on this have reportedly shifted since launch, so check Trojan’s own site for the current structure).

Trojan generates its own wallet for you. It’s not a custodian in the traditional sense, but that wallet still lives inside Trojan’s infrastructure. I couldn’t independently verify a specific third-party audit partnership, so I’m leaving that claim out. Here’s the concrete risk: if Trojan’s infrastructure is ever compromised, whatever sits in that generated wallet is exposed, not your main holdings elsewhere. Keep only trading funds in it. Move profits out regularly instead of letting them stack up in the bot’s wallet.

As of the most recent reviews I could find (mid-2026), Trojan has no publicly reported major security incidents, which is notable given how much volume it’s processing.

2. Dawn

Dawn’s whole pitch is solving a problem that anyone trading across more than one chain has run into — the annoyance of holding a separate wallet and a separate gas token for every network you touch. Run out of BNB and you can’t trade on BNB Chain, even if your Solana balance is fine.

Dawn gets rid of that entirely. You fund a single USDC balance, and when you buy something on a different chain, Dawn moves your funds there for you — no manual bridging. It supports Solana, Base, BNB Chain, Robinhood Chain, Arc, and Ink, and covers gas fees up to $5 per trade, so you’re not stuck buying small amounts of six different native tokens just to keep trading. The fee is 1%, and cashback is reportedly in the 25–55% range depending on activity.

On the wallet side, Dawn uses Privy for wallet infrastructure rather than storing your private key on its own internal servers directly, and the wallet can reportedly be exported. That’s a meaningfully different security model from a bot that generates and holds your keys entirely in-house.

The one real gap: no Ethereum or HyperEVM support yet. If you’re chasing a token on either of those chains, Dawn simply can’t execute the trade. You’d need a second bot open at the same time, which brings back some of the multi-tool friction Dawn is otherwise built to remove.

3. GMGN

GMGN works a bit differently from the others on this list — it’s as much a web-based analytics terminal as it is a Telegram bot, and its actual strength is smart-money wallet tracking. Instead of just showing you what’s trending, it tracks specific wallets over time and flags which ones are consistently profitable, which is what makes its copy-trading feature genuinely useful rather than a coin flip.

The fee is 1%, dropping to 0.9% with a referral code (MadeOnSol), though you’ll see third-party codes online advertising steeper discounts — I’d treat anything beyond the documented 0.9% as unverified until you’ve checked it yourself. GMGN supports Solana, Ethereum, Base, and BNB Chain,

and includes rug-check tooling that flags suspicious contracts before you trade, which matters given how many new launches are built to fail on purpose.

Here’s the specific way copy trading burns people: a wallet wins two trades in a row. You start mirroring it automatically. On the third trade, that wallet buys into a token that’s about to be rugged, and your money follows it in real time with no pause to think. A hot streak isn’t a strategy. Use wallet tracking as a lead to investigate, not a signal to auto-buy.

4. Maestro

Maestro has been a fixture in the sniping and copy-trading bot space for a long time, with support across Solana, Ethereum, BSC, and a few other chains, all at a flat 1% fee. It’s also one of two bots on this list with a real security incident worth knowing about.

In October 2023, an exploit in Maestro’s router contract let attackers pull roughly 280 ETH — around

$500,000 at the time — from users’ approved token balances. The team caught it, froze the router within about 30 minutes, and patched it. They then went further than a lot of projects do: Maestro refunded affected users using 610 ETH from its own treasury, worth over $1 million at the time, including a 20% bonus on top of what was actually lost.

That’s a genuinely good response. But it doesn’t erase what happened. A router-level vulnerability doesn’t care how polished the Telegram interface looks on top of it. Concretely: if a similar flaw ever resurfaced, you could see tokens or funds leave an approved contract without you initiating a transaction at all. Check Maestro’s current documentation for what’s changed in their contract design since 2023 before you assume today’s version carries the same risk as the one that got hit.

5. Banana Gun

Banana Gun built its reputation on speed — fast sniping, Anti-Rug protection, copy trading, limit orders, and DCA, at 1% on auto-snipe buys and 0.5% on manual trades.

It’s the second bot on this list with a documented incident. On September 19, 2024, attackers used what Banana Gun’s team later identified as a vulnerability in the Telegram message oracle the bot relies on, manually draining roughly $3 million from 11 users — notably, experienced traders rather than beginners, which is part of why the number of victims was small but the total loss was large. The team fully refunded affected users from its own treasury and added new safeguards afterward, including two-factor authentication for transfers and a two-hour transfer delay.

Worth noting: this wasn’t a platform-wide hack. It targeted specific high-value wallets, not every user on the bot. If you’re a small trader, that’s a somewhat different risk than Maestro’s router flaw, which could have hit any approved wallet regardless of balance. But if you build up a large balance in Banana Gun over time, you become exactly the kind of target this attack went after.

Banana Gun’s Anti-Rug filter is a genuinely useful precaution. The team hasn’t published the full algorithm behind it, so here’s the practical risk: a token can pass the filter and still turn out to be a rug pull, because the filter checks for known patterns, not every possible scam. Treat it as one extra check, not a green light.

Which Bot Actually Fits You

After going through all five, a handful of things mattered more than any feature list:

If you trade across multiple chains: Dawn’s single-balance, no-manual-bridging setup genuinely saves time and reduces how many places your funds are sitting at once. Nothing else on this list handles that as cleanly.

If you’re Solana-only and want depth: Trojan’s feature set — BOLT execution, copy trading, DCA, the Arena rewards layer — is built for someone trading heavily on one chain rather than spreading across several.

If your strategy leans on following other traders: GMGN’s wallet-tracking is the most developed version of that idea here, provided you treat it as a signal to investigate, not an autopilot switch.

If you’re using Maestro or Banana Gun: both are still widely used and both refunded their affected users in full, but go in knowing the history, keep balances small, and don’t treat a past refund as a guarantee about the future.

Across all five, the same rule applies: start with an amount you can afford to lose completely. Here’s why that matters concretely. If a token rugs, your funds are gone the moment liquidity is pulled, usually in seconds, with no recovery mechanism. If the bot itself is compromised, your funds can leave the wallet without you clicking anything. A bot only changes how fast you can act on a decision. It doesn’t protect you from a bad decision.

FAQ

Are Telegram trading bots safe?

Not fully. Two of the five bots here have had real security incidents where users lost real money. If a bot’s wallet is compromised, funds can disappear in minutes with no way to reverse it. Check the wallet custody model before you deposit anything.

Do I need a different bot for each chain?

Usually, yes. Dawn is the exception — one balance covers Solana, Base, BNB Chain, and a few others. With most other bots, if you want to trade a token on a chain the bot doesn’t cover with a shared balance, you’ll need to hold that chain’s gas token separately or run a second bot alongside it.

What’s the actual difference between these fees?

Most charge 1%, dropping to about 0.9% with a referral code. Banana Gun splits it: 0.5% on manual trades, 1% on auto-snipes. On a $1,000 trade, that’s the difference between paying $10 and paying $5.

Does a past security incident mean I should avoid a bot entirely?

Not automatically. Both Maestro and Banana Gun refunded every affected user in full. But a refund only happens after the fact — during the actual exploit, affected users had no way to stop funds from leaving their wallets in real time. Decide if you’re comfortable with that gap.

What’s the single biggest thing to check before using any of these? Whether the bot can export your wallet’s private key. If it can’t, and the bot’s servers ever go down or get compromised, you may have no way to recover funds sitting in that wallet.

Disclaimer

This is not financial advice. Trading memecoins carries real, substantial risk regardless of which bot you use to execute trades — a bot can only affect the speed and mechanics of execution, not the quality of the token itself. Fee structures, supported chains, and security postures change over time; verify current details directly with each platform before depositing funds. Every tool here is only as good as the judgment of the person using it.

I’ve Analysed the Best Telegram Trading Bots for Memecoins in 2026 was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

I compared ETH funding across Kraken, Coinbase, OKX and Hyperliquid. The rates looked wildly different until I checked what each number was actually measuring.

Kraken was showing an ETH funding rate almost 39× higher than Coinbase CDE when I checked both within the same three-minute window.

Disclosure: AI assisted with research, drafting and editing. The market observations, screenshots and calculations in this article were collected by me on September 29, 2026.

At first, I thought I had found a huge discrepancy.

Then I put the rates on the same clock.

Coinbase CDE became 0.0064% over eight hours. OKX was 0.0100%. Hyperliquid was approximately 0.0104%.

Kraken was still 0.249376%.

That was when the comparison became more interesting.

The problem was not simply that four exchanges displayed different percentages. The contracts, funding intervals and even the meaning of the displayed numbers were different.

Before deciding that one rate was “high,” I had to work out what I was actually comparing.

Four screens, four numbers

Here is what I captured during the short observation window:

Kraken — ETH Perp
0.031172% / hr · hourly

Coinbase CDE — nano Ether perpetual-style futures
0.0008% · hourly

OKX — ETHUSD UM X-Perp
0.01% · 8 hours

Hyperliquid — ETH-USDC Perp
0.0013% · hourly payment cycle

At first glance, these look like four versions of the same metric.

They are not.

The first problem becomes obvious as soon as the funding periods are written next to the percentages.

Method note: I captured the four public venue interfaces between approximately 09:50 and 09:53 UTC on September 29, 2026. I did not place a trade. I recorded the displayed rates first, then checked each venue’s funding interval and contract documentation before normalizing the rates.

Why crypto funding rates are easy to compare incorrectly

A rate quoted for one hour cannot be compared directly with a rate quoted for eight hours.

The first mistake I made was comparing percentages before comparing their clocks.

Coinbase says its US perpetual-style futures funding rate is calculated hourly, based on the relationship between its futures and spot prices.

OKX states that funding for its X-Perps is settled every eight hours.

Hyperliquid uses another structure. Its documentation describes an eight-hour funding-rate formula, while funding is paid every hour at one-eighth of the computed rate.

That means the raw percentages need a common clock before they can be compared.

Using an eight-hour equivalent:

• Kraken: 0.249376%
• Coinbase CDE: 0.0064%
• OKX: 0.0100%
• Hyperliquid: approximately 0.0104%

Something interesting happens after normalization.

OKX and Hyperliquid, which initially looked different because of the way their rates were displayed, land very close to each other.

Kraken still does not.

And that led to a second problem.

Same ETH does not mean the same contract

The underlying asset was ETH in all four observations.

The derivatives were not identical.

Coinbase CDE’s product is a US perpetual-style futures contract with its own clearing and funding process. Coinbase says the funding rate is calculated hourly, while the resulting cash adjustments are processed through its clearing system.

The OKX screen I captured showed an X-Perp, a long-dated derivative designed to behave similarly to a perpetual contract while retaining a distant settlement date.

Hyperliquid uses its own perpetual structure, funding formula and margin system.

Kraken has its own contract specifications and funding mechanics as well.

So this sentence:

“ETH funding was X%.”

is incomplete.

The better questions are:

Which ETH contract? On which venue? Over what funding interval?

The same asset name does not make two derivative products economically identical.

The 39× gap was real

The cleanest comparison in my screenshots was Kraken versus Coinbase CDE because both were showing hourly funding rates.

Kraken:

0.031172% / hour

Coinbase CDE:

0.0008% / hour

Dividing one by the other:

0.031172 ÷ 0.0008 ≈ 38.97

So, in those observations, Kraken’s displayed hourly rate was roughly 39× the Coinbase CDE rate.

That calculation is real.

What it does not establish is that Kraken is always 39× more expensive, or that the two contracts should normally have the same funding rate.

The observation describes a moment.

It does not create a permanent relationship between the venues.

The 273% headline that the same screen undermined

The Kraken number becomes even more dramatic if I annualize it.

0.031172% × 24 × 365 ≈ 273%

That would make a very clickable headline.

It would also be extremely easy to misunderstand.

It does not mean a trader should expect to pay or receive 273% over the next year.

It means something much narrower:

If that single hourly rate somehow remained unchanged for an entire year, its simple annualized equivalent would be roughly 273%.

The same Kraken screen gave me a reason not to treat that annualized figure as a forecast.

Alongside the current funding rate of:

0.031172% / hr

the interface showed a next funding rate of 0.0064350% / hr.

The next displayed rate was roughly 79% lower.

So a single screen contained both:

• a current rate capable of producing an eye-catching annualized number;
• and a next displayed rate already far below it.

A mathematically correct annualization can still create an economically misleading impression.

The same problem appears when a trading strategy is tested on historical data: funding rates, borrowing costs and liquidation mechanics need to be treated as explicit execution assumptions rather than background noise. Forvest’s guide to backtesting vs forward testing explains why those assumptions can change what a strategy appears to achieve once it moves from historical simulation to live conditions.

I checked Kraken again one hour later

I also ended up with a small follow-up check.

I captured Kraken once at approximately 08:50 UTC and again at approximately 09:50 UTC.

In both public-interface screenshots, the displayed current funding rate was:

0.031172% / hr

and the next displayed rate was:

0.0064350% / hr.

That does not prove the rate normally remains unchanged for an hour.

Two observations are not a time series.

They only allow a much narrower statement:

In two Kraken screenshots taken roughly one hour apart, the same current and next ETH funding rates were displayed.

That distinction matters.

I can document what the interface showed twice.

I cannot use two screenshots to describe Kraken’s normal funding behavior.

Before sharing a funding screenshot, check three things

After doing this comparison, I would no longer start with the size of the percentage.

I would start with three checks.

1. What is the funding interval?

One hour?

Eight hours?

Something else?

A percentage without its funding interval is incomplete information.

Two rates can look dramatically different before normalization and much closer afterward.

2. What contract are you actually looking at?

“ETH funding” sounds like a single market-wide number.

It is not.

The venue, contract structure, settlement design, collateral and calculation methodology can all matter.

Even products designed to behave similarly can have different mechanics underneath them.

3. Is the number current, next or annualized?

Those labels answer different questions.

A current funding rate describes the rate associated with a particular funding period.

A next or predicted rate refers to another period.

An annualized rate is a mathematical transformation of a shorter-period rate.

It is not automatically an expected annual return.

And none of those numbers, by itself, predicts ETH’s next price move.

The useful number was not the biggest one

I started this exercise thinking the interesting result would be:

Which venue has the highest ETH funding rate?

That turned out to be the wrong question.

The more interesting finding was how easy it was to build a bad comparison from four legitimate screenshots.

Same underlying asset.

Similar-looking funding labels.

Different clocks.

Different derivative structures.

Different meanings.

The 39× gap was real between the two observed hourly rates.

But understanding what that gap meant required more work than dividing two percentages.

That is what I would check first the next time I see a funding-rate screenshot shared as evidence.

A funding-rate screenshot is not a comparison until you know what the number is measuring.

Educational analysis of public derivatives-market data. No trades were placed for this comparison.

FAQ

Are crypto funding rates directly comparable across exchanges?

Not always. Funding intervals, contract design, collateral, price references and settlement mechanics can differ. The interval should be normalized first, and the underlying contracts should be checked before drawing a conclusion.

Does a high ETH funding rate predict ETH’s next move?

Not by itself. Funding describes conditions inside a derivatives market for a particular period. It can provide useful context about positioning and contract pricing, but a single funding-rate observation is not a directional price forecast.

Crypto Funding Rates: I Found a 39× ETH Gap in Three Minutes was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.

CoinGlass, Hyblock and CoinMarketCap describe different inputs. Here’s how to check the data before trusting a price-target claim.

Imagine a Bitcoin liquidation heatmap labeled 10:00, with a bright band at $90,000. At 10:15, a price chart reaches that level. Put the images side by side and the verdict seems obvious: the heatmap called it.

Disclosure: AI was used for research and drafting. Mobina Ebrahimi works in research and SEO at Forvest. Documentation checked September 28, 2026. The example above is hypothetical, not a market observation.

Two matching numbers are not yet a tested prediction. First, you need to know what produced the band, whether it was visible before the move, and what actually happened afterward.

The providers’ own documentation gives a useful starting point: similar-looking heatmaps can represent different evidence. That difference matters before you turn a colored zone into a trading decision.

Start with the data — not the color

The heatmaps discussed here map potential liquidation levels over time. They concern prices at which leveraged positions could be forcibly closed, not simply a list of liquidations that have already happened.

CoinGlass Model 1 starts with a calculation. Its aggregated heatmap documentation describes levels calculated from market data and liquidation leverage levels. The description does not establish that each band represents individually observed accounts. This is a statement about that endpoint, not every CoinGlass product.

Hyblock makes an estimation limit explicit. Its heatmap guide describes calculating levels and grouping them into price buckets. It warns that the tool predicts where levels open, but not where they close, and advises reading the displayed size relatively. A bright zone is therefore not a verified inventory of positions still open at that moment.

CoinMarketCap describes a position-based history. Its Bitcoin heatmap FAQ says it records liquidation levels from Hyperliquid positions on-chain roughly every five minutes. The page identifies Hyperliquid as its available coverage. It also says brightness is relative to the visible range, rather than an absolute dollar scale.

These are the providers’ descriptions, not an independent audit of their data. The comparison does not rank their accuracy. It establishes why “liquidation heatmap” is not the name of one universal dataset.

A model is not automatically useless. A position-based chart is not automatically complete. Neither becomes a forecast because its brightest band looks convincing.

A real position can still have a changing liquidation level

Observed positions answer a question about the input. They do not remove the conditions governing liquidation.

Hyperliquid’s documentation says liquidation becomes possible when account equity falls below maintenance margin — the minimum required to keep positions open. It uses a mark price, a reference price that can differ from the latest trade on the order book.

The same document explains that funding payments, or changing gains and losses on other positions in a cross-margin account, can alter the liquidation price. In cross margin, positions share collateral.

That matters when someone points to a candle touching a heatmap band. Was it the right venue and the right price reference? Did the relevant positions and account conditions still exist?

CoinMarketCap’s FAQ also distinguishes liquidation levels from orders already resting on an order book. A mapped level is not an executed liquidation. Reading it from a real position does not settle when — or whether — that event will happen.

The timestamp test: what was visible before the move?

Return to the hypothetical $90,000 band. The two screenshots invite a conclusion, but three checks separate a persuasive picture from supporting evidence.

Before the move: establish when the heatmap was captured, what data it contained, and which settings were selected. A historical chart generated later can show earlier times without proving what a reader could see at those times. That is a difference in evidence, not an accusation that a provider changed its history.

At the level: separate a price touch from a liquidation event. A claim that positions were liquidated needs an event record for the relevant venue and time. Two pictures displaying $90,000 cannot supply that record by themselves.

Across attempts: a single apparent success does not establish a dependable rule. Define the target, deadline and selection rule before checking the outcome, then include the cases that missed — not just the screenshot that worked.

A contemporaneous screenshot can support what was displayed. An event record can support what happened. Testing a forecast requires a defined rule and more than one selected outcome. Those are different jobs.

Give the screenshot a three-line record

You do not need to build a trading model to ask better questions. Before saving or sharing a heatmap, add three short notes:

Source: provider, exact chart or model, asset and venues. Is the input estimated or position-derived? One venue’s chart is not the whole Bitcoin market.

Time: capture time and time zone, latest data time if available, lookback, filters and legend units. Was this captured then, or generated later? Leave missing details unknown.

Claim: what are you saying the image demonstrates — a concentration of mapped levels, an actual liquidation, or a prediction? Name the extra evidence needed to support that claim.

For the opening example, even a genuine 10:00 capture might support only this: “The chart displayed a band near $90,000 before the later price touch.” That is a narrower — and more defensible — statement than “This chart reliably predicts Bitcoin.”

The same distinction between an original source and somebody’s interpretation runs through Forvest’s guide to analyzing crypto news. Here, the caption attached to a chart deserves as much scrutiny as the chart itself.

A useful map does not have to be a forecast

A heatmap can help locate concentrations within its stated method and coverage. It gives you a specific question to investigate. That is useful without pretending the display tells you where to buy or sell.

The bright band in our example has not become more or less colorful. What has changed is the claim we can responsibly make about it: first identify the input, then establish the timing, then check the outcome.

Knowing where a position could be liquidated is not the same as knowing where Bitcoin will go.

Educational analysis of data interpretation, not a recommendation to open, close or size a trade. Leveraged trading can result in substantial losses.

Bitcoin Liquidation Heatmaps: What the Bright Bands Really Show was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.