Turn Claude Code into an onchain research agent

Reading a block explorer used to be one of the highest-ROI skills in crypto. Following a funding trail, decoding contract calls, and spotting insider buys took real practice, and the people who could do it had an edge. That edge now costs a Claude subscription and access to blockchain RPC. The skill now is knowing what questions to ask. Where did the money come from? Who holds the admin keys? What can the dev actually do? An agent can answer all three, if you give it the chain and the question.

What does a research agent need?

Two things: a coding agent and a way to read the chain. If you followed our guide to set up Claude Code with Alchemy, you already have both. If not, install the Alchemy CLI and sign in:

alchemy install skills teaches Claude Code how to use the CLI, so it queries balances, transaction history, and contract state without guessing at commands. If you prefer tools over a CLI, the Alchemy MCP server exposes the same data as 168 tools across 100+ networks, and the Alchemy plugin for Claude Code bundles both in one install. The CLI can even pay for reads per call over x402, so an agent with a funded wallet doesn't need an API key at all.

Two optional additions make the agent a sharper investigator. An Etherscan API key lets it pull verified contract source code, so it can read what a function actually does instead of inferring from the selector. And Arkham's intelligence platform maps addresses to real-world entities, which turns "funded by 0x4f2a..." into "funded by a Binance hot wallet." Neither is required. RPC alone answers most launch questions.

What does a real investigation look like?

Take a real example from the video: a memecoin that launched through a Uniswap v4 hook with a bonding curve. The team was anonymous, and the launch was billed as fair. A fair-launch claim is checkable, so we checked it.

The prompt handed to Claude Code, lightly cleaned up:

This token launched via a Uniswap v4 hook with a bonding curve and claimed a fair launch. Find the hook, the token deployment, and the initial liquidity transactions. Check whether the deployer bought anything before the public could. Trace where the deployer wallet was funded from. Use the Alchemy CLI for RPC and data.

From there the agent worked the way an analyst would, just faster. It pulled the deployment transaction, identified the deployer address, reconstructed the first minutes of trading from transfer history, and followed the deployer's funding trail backwards. A few minutes later it came back with a verdict, and every claim in it carried a source: the transaction hash, the contract address, or the block it read.

The verdict on this launch had two sides. Mechanically, the fair-launch claim mostly held up. No presale, no team allocation, no hidden mint. But the deployer front-ran their own launch, buying in before the public had a chance. And the deployer wallet was funded through a privacy protocol, so the identity trail stops cold there. A launch can be mechanically fair and still come from someone who bought first and hid who they are. That nuance rarely survives a quick scroll through an explorer.

Nobody read an explorer page by hand to get there. The whole run was one prompt and a few minutes of waiting.

What questions should you ask?

The same questions work on every token you will ever look at, which is what makes them worth learning. Start with these:

• Where did the money come from? Trace the deployer's funding. Fresh wallet funded from an exchange is normal. Funded through a mixer or privacy pool is a signal worth weighing.
• Who holds the admin keys, and what can they do? Have the agent read the contract and list every privileged function: minting, pausing, upgrading, fee changes, blacklists. "What can the dev actually do" is a question about code, and the code is public.
• Did insiders move before the public could? Reconstruct the launch block by block. Deployer buys, sniper wallets funded from the same source, liquidity pulled and re-added.
• Where does the liquidity actually sit? Who provided it, whether it's locked, and what happens to holders if it leaves.

Here's a prompt you can paste into Claude Code to run the same investigation on any token. It asks for your target first, so it never invents one:

You can push further from there. Ask the agent to pull the verified contracts and walk you through the risky functions, or to diff the token against a standard ERC-20 and explain every deviation. The 12 things you can do faster with the Alchemy CLI post has more patterns worth stealing.

Start asking better questions

All of this runs on the free tier. Install the Alchemy CLI, sign in, and your agent is reading mainnet within a minute. No contracts, no waitlist, no API key if you route reads through x402. And when your research agent graduates into something that acts on what it finds, our guide to building onchain agents covers the wallet, payment, and data layers that come next.

Block explorers aren't going anywhere, and knowing your way around one still helps. But the reading is now the agent's job. Deciding what to ask is yours.