Etherscan has rolled out a suite of tools that allows AI agents and coding assistants to pull live blockchain data directly from the explorer's API. The release targets developers building agents that need verified onchain facts across more than 60 EVM-compatible networks. The company announced the "Build with AI" launch in a post on X on August 31, 2026.

What Etherscan actually shipped

The release bundles three ways for a machine to reach Etherscan's data. The first is a Model Context Protocol (MCP) server, a hosted interface that agents such as Claude and ChatGPT can call in plain language. MCP is an open standard, introduced by Anthropic in late 2024, that standardizes how AI applications connect to external data sources, and it has since been adopted across major AI assistant platforms. The second is a command-line tool for terminals, scripts, and CI pipelines. The third is a set of installable agent "skills."

According to Etherscan's documentation, the MCP server runs at a single endpoint, mcp.etherscan.io/mcp, and exposes 20 tools. Those cover the basics of reading a chain: native and token balances, normal and internal transactions, transaction and receipt lookups, contract details, gas prices, and event logs. A single connection reaches every supported network.

The Command-Line Interface (CLI) follows the API's structure, with commands in the form etherscan <module> <action> and JSON output by default, plus table and CSV options for human readers. Everything authenticates with a standard Etherscan API key, and a free key works across every supported chain. That key system has existed for years: Etherscan's public API has long been a building block for wallets, portfolio trackers, and analytics dashboards, and the new tools expose the same underlying data to machine callers.

Skills that trace money and read contracts

Etherscan's public GitHub repository lists four installable skills, added through an npx skills add command or by copying a folder into an agent's skills directory. Etherscan Flow traces money movements across addresses and writes them into a case file. Etherscan Contract Review explains what a verified deployed contract does. A transaction debugger reconstructs what a given transaction did and why. An orchestrator skill routes a task to the right interface.

According to the repository, the stated point of the skills is that "every address, amount, and transaction hash comes from a live API call, never invented." The whole suite is read-only and respects the API's normal quotas and rate limits.

Fighting the hallucination problem

The release is viewed as a response to the hallucination problem that has spread into crypto-AI tools. Language models can fabricate a wallet balance or a transaction that never happened, and either error can be costly. This typically happens when models lack the right data — in most cases because they cannot crawl the pages containing the information they need.

In its documentation, Etherscan also warned users not to paste their API keys into fake or fraudulent MCPs posing as legitimate services. The company stated that only one official MCP server exists and cautioned that listings on MCP marketplaces calling themselves "Etherscan MCP" are unaffiliated.

Built for an internet where bots rival people

The scale of usage and agentic activity calls for appropriate safeguards to prevent lapses. In early June, AI agents surpassed humans as the internet's largest source of traffic. The shift was confirmed by Cloudflare CEO Matthew Prince, who said it happened sooner than he had predicted. Cloudflare Radar data later put agentic bots at 57.4% of web traffic versus 42.6% for humans, with North America skewed further toward bots at 68.6%. Humans have since retaken the lead, but bots still account for more than 35% of web traffic over the past four weeks.

Publishers, for their part, are now working to make their content easier for AI agents to access. Etherscan's release fits the same pattern: it is publishing machine-readable documentation and a query layer built for software readers, not just human ones. For developers, the practical effect is that agents can be pointed at a single authenticated endpoint rather than scraping explorer pages, and future additions to the tool set would follow in Etherscan's documentation and GitHub repository.