Gemini 推出 Agentic Trading：讓 Claude、ChatGPT 直接下單受監管交易所，執行策略交易

Gemini Officially Launches Agentic Trading Feature, Becoming the First Regulated U.S. Cryptocurrency Exchange to Openly Integrate AI Agents for Automated Trading. The platform is based on the Model Context Protocol (MCP) open standard, allowing AI models like Anthropic Claude and OpenAI ChatGPT to execute buy and sell orders directly through APIs on behalf of users. The initial release includes three modules covering real-time market data, spread analysis, and historical candlestick data.
(Background recap: Coinbase launches Agentic Wallets: enabling AI agents to autonomously trade, pay, and earn, integrated with x402 protocol)
(Additional background: IPO disappointment! Gemini faces a class-action lawsuit for “misleading investors,” stock plummets 80%, and 25% of staff are laid off, exiting multiple countries)

Table of Contents

Toggle

• MCP Architecture: Turning the entire trading API into a language AI can understand
• Gemini’s Bet: AI handles execution, humans focus on strategy
• Not Just Gemini: AI × Crypto bridging protocols are advancing in multiple directions

Gemini announced on its official blog that Agentic Trading is now live—this is the first regulated U.S. crypto exchange to offer a feature that allows AI agents to directly execute trading operations. Unlike traditional API integrations that require manual confirmation for each command, this system grants AI models full access to the exchange’s functions, enabling autonomous completion of tasks from market queries to order placement.

MCP Architecture: Turning the entire trading API into a language AI can understand

Gemini chose to build this system based on the Model Context Protocol (MCP). MCP is an open protocol led by Anthropic, designed to enable AI models to interface with external tools and services through a unified API. By integrating its entire trading API into MCP, Gemini ensures that any MCP-compatible AI model—such as Anthropic Claude and OpenAI ChatGPT—can directly access the full suite of exchange functions.

The initial release includes three modular Trading Skills (pre-built functional modules): Get Market Data for real-time price information, Find the Spread for analyzing buy-sell spreads, and Retrieve Candles for historical candlestick data. The strategy coverage ranges from basic order placement to complex multi-leg positions.

Gemini’s Bet: AI handles execution, humans focus on strategy

Gemini explicitly states in its announcement that this is not merely a feature launch but a redefinition of trading paradigms: “We believe this is the beginning of a fundamental shift in how people interact with financial markets,” and adds, “Agentic trading is not just a feature—it’s a new paradigm—AI handles execution, modes, and discipline; you focus on strategy and goals.”

The timing is noteworthy: Gemini only announced layoffs of 25% in February this year, simultaneously exiting markets in the EU, UK, and Australia to concentrate resources on the U.S. domestic market. Now, with the high-profile return of Agentic Trading, it clearly signals: abandoning geographic expansion in favor of AI agent infrastructure to capture the next product cycle’s commanding position. The stock reaction was modest, with GEMI rising only 0.25% to $4.40 on the day, up 7% over the past month, but down 55% since the start of the year.

Not Just Gemini: AI × Crypto Bridging Protocols Are Advancing in Multiple Directions

The competition in this space has quietly taken shape. Coinbase’s incubation project, now under the Linux Foundation, x402 protocol, also targets on-chain payments and interactions for AI agents; Stripe-backed Tempo Network has launched the Machine Payments Protocol, focusing on automated settlement scenarios between AI agents.

As an open standard led by Anthropic, MCP is chosen by Gemini as its backbone rather than developing a closed protocol. This decision makes Agentic Trading inherently compatible with all MCP-supported AI models on the market, allowing ecosystem expansion at a much faster pace than closed solutions. Whether AI agents can truly become the mainstream interface for retail and institutional order placement may be a pivotal point in 2026.