Coinbase’s layer-2 blockchain enables Claude and ChatGPT to execute transactions with user confirmation
Base, the Ethereum layer-2 blockchain operated by Coinbase, launched a tool on May 27, 2026 that connects Base accounts to artificial intelligence agents, allowing models like Anthropic’s Claude and OpenAI’s ChatGPT to execute blockchain transactions on user wallets after explicit confirmation.
The Base Model Context Protocol (MCP) tool enables AI agents to transfer funds, swap tokens, check balances, and review transaction history across supported protocols including Morpho, Moonwell, Uniswap, Aerodrome, Avantis, Bankr, and Virtuals. The mechanism preserves user control: agents propose actions in chat, the Base wallet opens in a separate window, and users must actively confirm or cancel each transaction. Agents do not access private keys, and every transaction undergoes the same review flow as standard Base account requests, with asset changes simulated before confirmation.
“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,” said Lincoln Murr, head of AI Product for Coinbase. Murr also described the tool as a “nice wrapper” on top of APIs.
The launch accelerates adoption of the Coinbase x402 protocol, an agentic AI payment standard introduced in May 2025. Over the past 30 days, x402 processed $1.1 million in transaction volume. Industry proponents argue agentic payments represent a critical use case for crypto because AI models face barriers accessing traditional banking systems and require digital assets to transact independently.
The integration includes support from Fireblocks, which provides agentic payment infrastructure, and Socket, a developer platform. Base MCP enables what the ecosystem terms a “micro-transaction economy” where agents can execute small-value payments in crypto without human intermediation beyond the confirmation step.
Security concerns accompany the expansion. Research from Google and leading universities argues AI agents should be treated as untrusted system components, a stance that contrasts with the trust model Base MCP requires for transaction confirmation. Socket separately discovered malware targeting crypto developers by injecting hidden instructions to hijack AI coding assistants, raising questions about AI agent integration risks in development workflows.
Technical and Operational Details
Base MCP operates within the existing Base account infrastructure. Every transaction follows identical review protocols as non-agentic wallet operations, ensuring consistency with Coinbase’s established security posture. The tool does not grant agents autonomous signing authority; confirmation remains mandatory.
The exact launch date within May 27 was not specified in available sources. Geographic availability and specific technical differences between Base MCP and alternative agentic wallet designs were not detailed. Socket did not name the malware variant or quantify its scope beyond crypto developer targeting.
Market Context
Agentic AI payments remain nascent. The $1.1 million volume across x402 in the prior 30 days reflects early-stage adoption. Base’s integration of MCP represents one of the first major blockchain infrastructure plays to embed agent transaction support directly into a layer-2 ecosystem, positioning Coinbase’s platform as a potential standard for agent-wallet interaction as AI models become more economically autonomous.
