In the past, when we used AI for trading analysis, we needed to copy the results from the tool and then switch to the exchange to place orders manually. On the surface, it just added a few more steps, but in a rapidly changing market, those few seconds spent switching tools often meant that prices had already changed.

More critically, this division brought not just an efficiency loss. AI's analysis was on one end, while trade execution was on the other, making it impossible to truly connect the two. The agent could perceive trends and generate strategies, but once it came to actual trading, manual intervention was still required to complete the operation. This is clearly not the "intention era" we were expecting, nor the ideal "intention trading."

The Agent Trade Kit launched by OKX is designed to completely tear down this wall.

What is the Agent Trade Kit?

The Agent Trade Kit is an open-source AI trading toolkit launched by OKX, with a single core goal: to allow the AI agent to connect directly with the OKX exchange, completing the entire process from analysis to order placement in one conversation, without any manual intervention in between.

It supports mainstream AI clients available on the market today, including Claude, OpenClaw, Cursor, VS Code, and all tools compatible with MCP protocols. No matter which AI tool you normally use for market analysis and judgment, once integrated, you can directly have it execute trades for you.

The Agent Trade Kit is now live on npm, offering two versions: okx-trade-mcp (MCP server, suitable for AI client users) and okx-trade-cli (CLI tool, suitable for developers and quantitative traders).

What can the Agent Trade Kit do?

Currently, the Agent Trade Kit is the most comprehensive MCP product covering functions in the market. With 83 tools and 7 major modules, it supports five major trading categories: spot, perpetual contracts, delivery contracts, options, and leverage, covering the complete trading lifecycle from market data retrieval to strategy execution.

• Market Data: Real-time quotes, K-line data, funding rates, mark prices, positions, limits on price increases or decreases... This information used to need to be checked on various pages by oneself, but now it can be retrieved directly in AI conversations all at once, allowing for seamless analysis without needing to switch pages back and forth.
• Spot and Contract Trading: Spot order placement, going long or short on perpetual contracts, and trading delivery contracts—simply tell the agent in natural language what you want to do, and it will help you construct the order, find the best price, and complete the execution. Operations that used to require you to watch the board and click manually can now be completed with a single sentence.
• Options Trading: This is one of the most unique capabilities of the Agent Trade Kit. Currently, it is the only MCP toolkit in the market that supports AI Agent access to real-time options markets; other exchanges do not yet offer similar support. Hedging positional risks, taking directional profits with limited losses, and building multi-leg options strategies can be achieved only with the Agent Trade Kit.
• Algorithmic Orders: Set OCO orders (orders to simultaneously take profits or stop losses), trailing stops, and condition triggers using natural language. You no longer need to study how to configure the platform's algorithmic interface; just tell the agent your intent, and it will translate it into specific order parameters and execute them.
• Strategy Trading: Configure grid strategies and DCA (Dollar-Cost Averaging) strategies through conversations. Grid strategies are suitable for automatically buying low and selling high in volatile markets, while DCA strategies are appropriate for gradually building positions over the long term. Previously, configuring these required filling in parameters one by one in the exchange interface; now, just describe your strategic ideas, and the agent will set it up for you.
• Account Management: Directly query your account balance, positions, transaction history, fee rates, and position mode settings in AI conversations… Always have a full view of your account without needing to log in to the exchange separately.
• Simulated Trading Mode: A complete simulated trading environment is built in, and all 83 tools are available in simulation mode. Test strategy logic with real market data, verify the agent's execution effects, and confirm everything is fine before switching to real trading. For users just starting to trade with AI, this feature significantly reduces trial and error costs.

Two Access Methods, Choose as Needed

As mentioned earlier, the Agent Trade Kit provides two versions:

• MCP Server (recommended for beginners and heavy users of AI tools): A local service compliant with the Anthropic MCP standard, seamlessly integrated with mainstream AI clients like Claude Desktop and Cursor. Once configured, you can complete all trading operations directly in the AI conversation window in natural language, without needing to know any code, resulting in a very low entry barrier.
• CLI Tool (recommended for quantitative traders and developers): An independent terminal command-line tool that supports shell pipes, cron scheduling tasks, and script integration, allowing it to run independently without relying on any AI clients. Compared to MCP, the CLI version consumes fewer tokens and executes faster, making it more suitable for users needing high-frequency operations or automated scripts. The same toolkit is available in two usage methods, allowing users to pick what they need.

In addition, the Agent Trade Kit also provides four plug-and-play Skills modules: Market Data, Trade Execution, Portfolio Management, Strategy Trading. If you don’t need the complete toolkit, you can just load the modules you need, allowing the agent to be more focused and responsive.

Examples of Application Scenarios for the Agent Trade Kit

The Agent Trade Kit is tailor-made for all trading scenarios, covering spot, contracts, grid, Martingale, algorithmic orders, arbitrage, options, and more. With just one sentence, you can build any trading strategy and also capture real-time market data for analysis, forming a complete closed loop of "market retrieval → intelligent analysis → strategy generation → automatic execution."

For example, if you want to make a limit order, you can directly tell the agent, "Go long on BTC at market price for 0.1, take profit at 92,000, and stop loss at 84,000," completing the order placement and simultaneously setting the take profit and stop loss in one sentence; if you want to arbitrage, you can let the agent monitor the funding rate in real-time and automatically trigger order placements when the price difference reaches the target, without needing to watch the board; for long-term positions, you can set target asset ratios, and the agent will automatically calculate and execute position rebalancing. Whether it is for short-term executions or long-term allocations, just describe your intent, and leave the rest to the agent. Here are some typical usage scenario examples:

What has the Agent Trade Kit done for AI safety?

For many people, handing over API keys to AI tools is their biggest concern. If the keys leak, the risk of asset loss arises. Therefore, the Agent Trade Kit has put considerable effort into security design, with one core principle: your keys are always only on your own device.

• Keys are stored locally only, inaccessible by AI during the whole process. API keys are only saved locally, with signatures completed locally by MCP using HMAC-SHA256. What is sent to AI is only the "trading intent," without any sensitive information. Even if conversations are intercepted, operable credentials for accounts cannot be obtained.
• Permission isolation and read-only mode. The system automatically reads API permission configuration upon startup. If trading permissions are not enabled, the order placement tools will not appear in the agent's list; in read-only mode, AI can only query data and cannot execute trades, thus fundamentally preventing unauthorized operations at the architectural level.
• Module-level permission control. You can precisely decide which functional modules the agent can use, limiting its operational scope and avoiding unnecessary risk exposure.
• Complete isolation between real and simulated accounts. A fully simulated environment is physically isolated from real accounts. Testing strategies will not mistakenly touch real assets.
• Rate limiting and anomaly protection mechanisms. The system has built-in rate limiting and abnormal instruction interception mechanisms to prevent abnormal calls or erroneous operations from amplifying risks.
• Fully open-source and auditable. The Agent Trade Kit is open-sourced under the MIT license, with all code being publicly transparent. There is no need to rely on trust; every line of logic can be independently reviewed.

Of course, while tool security is fundamental, good usage habits are equally crucial. The Agent Trade Kit addresses the issue of "keys not being exposed to AI models," but the inherent limitations of AI still exist. In extreme market conditions, the model may also make decisions that do not align with your expectations. Handing over execution to AI does not mean relinquishing judgment rights. The logic of every transaction remains your responsibility.

Therefore, it is recommended to set up a separate sub-account for AI use and employ a dedicated API key, strictly isolating it from the main account and assets, placing only funds for AI operations; do not send sensitive information such as API keys, private keys, or mnemonic phrases in any conversation; set trading permissions to "need approval" to ensure critical operations receive confirmation; during the first use, prioritize validating the process in a simulated environment before switching to real trading; also, only use official provided or recognized skills to avoid enabling third-party plugins from unknown sources.

Only under the dual security mechanisms and usage regulations can AI truly become a tool for improving trading efficiency.

Let the Agent Truly Integrate into the Trading Market

The Agent Trade Kit is not an isolated tool, but rather an essential part of OKX's construction of AI-native trading infrastructure. Previously, OKX had opened AI capabilities for its on-chain autonomous Web3 development platform Onchain OS, covering key scenarios such as on-chain wallet management, DEX trading, and on-chain payments, enabling the agent to genuinely act independently within a Web3 environment. The Agent Trade Kit complements the AI capabilities for centralized trading, allowing AI to directly participate in mainstream trading scenarios such as spot, contracts, and options.

When the abilities of on-chain and off-chain, CEX and DEX are simultaneously connected, AI is no longer just a mere market analyst or strategy generator, but becomes a participant that can continuously act across multi-market environments.

This is not only an expansion of tool capabilities but a change in trading paradigms—from "humans using tools" to "humans defining intent, agents executing trades."

Under the full-scene coverage of CEX + DEX, an environment is forming where AI can operate collaboratively across markets. The agent is no longer just confined to the dialogue box, but is starting to truly integrate into the global trading network.

Start using

Agent Trade Kit: https://www.okx.com/zh-hans/agent-tradekit

Onchain OS: https://web3.okx.com/onchain-os

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。