How HypurrClaw Built an AI Crypto Trading Agent with the Blockscout API

AI agents are becoming one of the fastest-growing categories of Blockscout API users. HypurrClaw is a good example of why: it's an AI crypto agent that lets users research onchain activity and execute trades from a single chat thread, and it leans on Blockscout for multichain explorer data across every EVM chain it touches. We caught up with founder Sayo to talk about how the agent works, where Blockscout fits under the hood, and what people are actually doing with it.

Hi Sayo! Great to connect with Blockscout API users in the wild. Give us the tl;dr on HypurrClaw. What does it do and how does it work?

HypurrClaw is a personal AI crypto agent. You talk to it in natural language to research onchain activity, then execute when you're ready: swaps, perps via Hyperliquid, prediction markets, auto sell, auto take-profit, auto stop-loss.

Most users are on Telegram today at @HypurrClawBot. Paste a token, wallet, or transaction, ask a question, or place a trade. Users stay in control, and nothing moves until they approve it. Automations only run after the user authorizes the rules upfront.

The web app at hypurrclaw.xyz is live with chat, wallet, workspace, and automations. Native iOS and Android apps are in beta. We're building toward a unified crypto super app: wallet, portfolio, intelligence, swaps, perps, predictions, alerts, and auto trades in one place, with the agent on every screen.

Where does Blockscout plug in? Any other integrations, and how easy were they to configure?

Blockscout is our EVM intelligence layer. It gives us multichain indexed explorer data on demand: balances, transfers, decoded transactions, contracts, tokens, and address metadata.

Users hit it constantly without realizing it. When they paste a transaction, check a contract before a swap, or ask what's in a wallet, Blockscout powers the answer without anyone opening an explorer. That same layer matters on Polygon and other EVM chains whenever collateral, deposits, or settlement activity needs a plain-English onchain read.

We also use it to analyze high-performing wallets. You can trace what profitable wallets bought and sold, when they rotated, where funds moved next, and whether an address is a known entity like a protocol, CEX, or tagged counterparty. Paste a 0x address and ask "what has this wallet been doing?" or "is this smart money?" and Blockscout supplies the recent transactions, token transfers, balances, holder context, and metadata labels behind the answer.

On Solana we use Helius for indexed wallet and transaction data. On EVM, Blockscout is that foundation. Beyond those two, the main third-party piece is Privy for wallets and signing. Blockscout was the easiest to integrate with a single API key, multichain coverage, and endpoints that map to how people actually ask EVM questions.

Why did you pick Blockscout over other API or MCP options?

There were a few reasons.

1. Multichain was a big draw. Base, Ethereum, Arbitrum, BSC, Polygon, and more without juggling per-chain explorer APIs.
2. Indexed and decoded context together in the API. Transactions, contracts, holders, and labels come out of one feed.
3. It matches how users talk. People paste a 0x hash or name a chain, and Blockscout resolves it.
4. Wallet forensics at scale. There's enough indexed history to spot patterns in profitable traders: early token buys, rotation timing, counterparty flows, and labeled addresses, all without stitching together five chain-specific APIs.

We didn't need a separate explorer MCP per chain. Blockscout covered multichain search, decoded transactions, and address context in one place.

For prediction-market users, that matters twice. There are pre-trade contract and wallet checks on EVM chains, and there's onchain verification when someone asks "did my deposit land?" or "what happened in this tx?" after funding or redeeming.

What's been your biggest pain point building the app and the Telegram bot? Have you had any reliability issues along the way?

Receiving messages is easy. Shipping a bot that streams replies, keeps execution user-gated, keeps keys out of chat, and works in both groups and DMs is not.

The reliability pain has mostly come from third-party APIs rather than Telegram itself. Execution paths add their own layer: rate limits, wallet funding, and UX design where a pending approval means the user still has the tap, not that something failed. The principle stays the same throughout. Nothing executes unless the user chooses to proceed, including automations they authorized in advance.

How are people using HypurrClaw? Any surprising query patterns?

There's three main groups using the app. Traders run perps, brackets, auto sell, take-profit, stop-loss, and token scans. Researchers do wallet forensics, transaction explanations, and smart-money analysis, tracing a high-PnL address's buys, sells, and rotations through Blockscout activity and token transfers, plus prediction-market research like "what's trading on the Fed meeting?" or "show me sports markets with volume." Builders use the same research and trading stack without wiring it themselves.

On EVM, Blockscout shows up constantly: "what happened in this tx?", contract checks before swaps, "where did funds go?", "what is this wallet trading?", "follow the money from this profitable address."

On predictions, the pattern is different but familiar. People search or browse markets in chat, pull snapshots and orderbooks, then preview an order and place it on their terms when they're ready. Users don't want a separate odds tab and an explorer tab. They want one thread.

As for surprises: natural-language trades beat menus every time. People analyze random wallets across Solana and EVM in a single conversation. Users paste a bare 0x with no chain specified and expect us to figure it out. Sports and macro event queries usually start as plain English, not market IDs. And profitable-wallet rabbit holes are real. Someone drops one hot address and expects a full narrative of entries, exits, and current holdings.

What's next for HypurrClaw? More data sources, or more from Blockscout?

The goal is a crypto super app for AI-assisted trading: wallet, portfolio, intelligence, swaps, perps, predictions, and auto trade, with an agent bar on every screen. Web is live today, native iOS and Android are in beta, and builders can plug into the same stack at hypurrclaw.xyz.

With Blockscout, we're going deeper on it as the default EVM intelligence layer: portfolio, pre-trade checks, address intel, and smarter wallet-activity and profitable-trader research. That context feeds our EVM automations for auto sell, take-profit, and stop-loss.

Prediction markets are a big part of the product. How do they fit with Blockscout in practice?

They're complementary layers in the same conversation.

Prediction-market reads answer: what can I trade, what's the price, what's the book? That covers discovery, snapshots, orderbooks, and live order placement when the user is ready.

Blockscout answers: what actually happened onchain? That comes up when a user funds a wallet, moves collateral, redeems a resolved position, or pastes a Polygon tx hash and asks for a plain-English breakdown.

A typical flow has three steps.

1. Research. Paste a contract or ask about a coin, project, or market: "analyze this token," "what's the narrative," "who's buying," or for predictions, "what's trading on this election?" The agent pulls market data, wallet activity, odds, and orderbooks into one thread.
2. Verify. "Do I have collateral ready?" "What's in this wallet?" "Explain this 0x transaction." Blockscout grounds the answer in chain state, with balances, transfers, and decoded calls, without sending the user to a separate explorer.
3. Execute. Preview a swap, perp, or limit order. The user approves in Telegram or the web app, then Privy handles signing. No keys in chat.

That's also what we'll integrate into the super app: predictions as a first-class surface, Blockscout-backed EVM context one tap away when something onchain needs explaining, and the agent on every screen so users never bounce between an odds app and an explorer app.

Anything else? How can people get started?

Whether you're into prediction markets, perps, or memecoins, HypurrClaw gives you one agent and one thread: research, onchain context, and execution when you're ready. Users stay in control of every trade.

• Telegram: @HypurrClawBot. Ask for markets, perps, or a token, or paste a wallet or tx for a plain-English onchain read
• Web: hypurrclaw.xyz
• Docs: docs.hypurrclaw.xyz
• Community: @hypurrclaw on X

HypurrClaw is built by Sayo (@sayodotfun).

Building with the Blockscout API? Get a free key at dev.blockscout.com and tell us what you're making. We love featuring projects like this one.