Base, the Ethereum layer-2 blockchain from crypto exchange Coinbase, has launched a tool to connect Base accounts to artificial intelligence agents for blockchain operations.
Base said on Tuesday that its new Base MCP (Model Context Protocol) allows users to ask AI agents such as Anthropic’s Claude or OpenAI’s ChatGPT to transfer funds, swap tokens, check balances, review transaction history, and use supported apps within the ecosystem.
The tool lets users manage their crypto directly from an AI model’s chat interface and can also interact with crypto protocols such as Morpho, Moonwell, Uniswap, Aerodrome, Avantis, Bankr and Virtuals, said Base.
The AI agent proposes an action in the chat, then the Base wallet opens in a new window where users can confirm or cancel the transaction. The agent does not have access to private keys, and every action must be confirmed by the user.
Agentic payments have been touted as the next major use case for crypto, as backers argue AI models will have issues accessing the banking system and will need to use digital assets to transact.
Base said that every transaction the agent proposes goes through the same review flow users see for any Base account request, with asset changes simulated before a user confirms.
Lincoln Murr, head of AI Product for Coinbase, told Fortune that “unlike siloed agentic wallets that only live in a terminal, your Base Account travels with you — trades, history, and portfolio sync whether you’re in-agent or in the Base App.”
Base MCP will also expand the adoption of the Coinbase x402 protocol, an agentic AI payment standard the company launched in May 2025.
Related: Fireblocks launches agentic payment support, joins x402 Foundation
Murr described MCP as a “nice wrapper” on top of APIs. Together with x402, it enables a new micro-transaction economy where agents can make tiny payments in crypto.
However, that economy is currently in its early stages, with x402 processing just $1.1 million in volume over the past 30 days, according to x402scan.
X402 statistics over the past 30 days. Source: x402scan
The use of AI agents for crypto payments has its detractors, with a recent research paper from Google and leading universities saying that AI agents should be treated as an untrusted system component.
The researchers said that AI agents should clearly distinguish between instructions and untrusted data to avoid attackers duping the agent by hiding malicious instructions.
Earlier this week, the developer platform Socket discovered malware targeting crypto developers by injecting hidden instructions to hijack AI coding assistants.
Magazine: Polymarket seeks Japan entry, Harvard dumps entire ETH position: Hodler’s Digest
