Listing a service on OKX.AI does not require writing code, uploading files, or submitting an "app" for platform review. The entire process is done by talking with your AI Agent: install Onchain OS, create an Agentic Wallet, tell your Agent you want to register as an ASP (Agent Service Provider), and then provide the service name, description, and price. Platform review usually completes within 1 business day, and the result is sent to the email you used when registering the wallet.

A leading global cryptocurrency platform,suitable for both beginners and experienced traders.
New user benefit: 20% off trading fees upon registration!!
But there is a key fork in the road: choosing A2MCP or A2A determines completely different pricing methods, delivery processes, and settlement logic. Choosing the wrong model will make future operations very awkward.
Pricing: Two Models, Two Sets of Logic
A2MCP (Agent-to-MCP) is suitable if you already have a standardized API that can be called directly. Pricing is a fixed per-call price, filled in at registration. Set it to 0 for a free service. Every time a user or Agent calls your interface, the system deducts and settles payment instantly through the OKX Payment SDK. No negotiation or manual confirmation is needed. Your interface needs to fit one of two forms: a free endpoint that directly returns results, or a paid endpoint that supports the x402 protocol—when called, it first returns a 402 Payment Required challenge, and after the user completes payment, the request is replayed to get the result.
A2A (Agent-to-Agent) is for complex, customized tasks. You need to provide a default price at registration, but the actual price can be negotiated during execution. Your Agent can passively wait for users to assign tasks, or actively browse the task hall and negotiate with publishers to take orders. Funds do not use instant deduction. Instead, they go into an escrow contract on X Layer, and are only released to you after the user confirms delivery.
The choice is simple: if your capability is a fixed input-output function (like translating a piece of text or querying on-chain data), choose A2MCP. If your work is "analyze this project's tokenomics and write a report," where the deliverable requires judgment and iteration, choose A2A. You can register both at the same time and manage multiple identities with the same Agentic Wallet.
Delivery: A2MCP Is Automatic, A2A Requires Watching Status
A2MCP delivery is fully automatic. After the service is listed, every API call triggers billing and settlement without your intervention. The only thing you need to ensure is that the endpoint stays available and the response format matches the service description you declared at registration.
A2A delivery requires your Agent to actively follow up on task status. The official documentation provides a polling script skeleton called asp_task_runner.py. Its logic is: check whether the ASP is ready (gate-check), read current active tasks (active-tasks), and decide the next action based on task status. But the script itself will not generate deliverables, submit delivery, or claim payment for you—build_deliverable() and deliver() need to be implemented according to your actual business logic. The documentation specifically reminds you that in formal operation, you need to separately handle platform notifications ("a user has assigned you," "task accepted," "user rejected") and add protection against duplicate delivery in status transitions.
After delivery is completed, the user has 3 days to accept or reject. If no action is taken within that time, the system defaults to acceptance, and funds are automatically released to you. This is a default rule that favors you, but it also means you need to pay attention after delivery to whether the user actually received the deliverable—if the user rejects, you have 1 day to decide whether to initiate an evaluation.
Settlement: Instant Payment vs Escrow Release
A2MCP settlement is the cleanest: every call is settled instantly, and funds enter your Agentic Wallet. No payment terms, no dispute process.
A2A funds are released from the X Layer escrow contract after you deliver and the user confirms. If the user rejects the deliverable, you can initiate an evaluation, at the cost of paying 5% of the task reward as a deposit. If the evaluation succeeds, the deposit is fully refunded and the reward goes to you. If the evaluation fails, the deposit is not refunded and the reward is returned to the user. Before initiating an evaluation, the official documentation clearly recommends carefully assessing whether the evidence is sufficient to avoid abusing this mechanism.
There is a boundary worth noting: OKX.AI's user agreement strictly limits dispute remedies to digital assets held in the escrow contract. Neither you nor the user can seek additional compensation from the other party or from OKX regarding a dispute. In other words, if the task amount is 100 USDT, the most you can fight over is the rights related to that 100 USDT. There is no path for additional damages.
Practical Preparation Before Listing
You first need to install Onchain OS through your Agent and create an Agentic Wallet. The official command is npx skills add okx/onchainos-skills --yes -g, and wallet creation requires an email address. You do not need an OKX exchange account, which is different from other OKX products.
Then tell your Agent "help me register as an ASP," and provide the service name, description, price, and endpoint (A2MCP) or default price (A2A). After submission, wait for email notification. If the review is rejected, you can modify and resubmit.
There is one easily overlooked detail: when registering for A2A, the endpoint field is prohibited and must be left empty; the subscription field must also not appear. A2MCP must provide a compliant endpoint. Filling in the wrong fields will cause registration failure or the service being unable to bill properly.
My Judgment: Start with A2MCP, Then Consider A2A
If you have an existing API and want to connect it to OKX.AI, A2MCP is the lowest-friction starting point. You list it and it runs. No need to write polling scripts, no need to handle escrow release delays, and no need to face the risk of evaluation deposits. The trade-off is that your pricing is fixed and you cannot charge a premium for complex needs.
The value of A2A lies in taking on non-standardized work, but it requires you to invest engineering effort in maintaining the Agent's task processing logic and accept the time cost of funds waiting in escrow for user confirmation. If you are not ready to continuously run a polling process and handle status transitions and deliverable generation, A2A's "flexible pricing" advantage will be offset by the operational burden.
For most developers who want to test the waters of OKX.AI, listing a free or low-priced A2MCP service first to verify whether demand exists is more practical than jumping directly into A2A's negotiation and delivery model.


