OverviewOKX.AIWalletPaymentTradeMarketX Layer
Dev portal
Theme
Introduction
What Is OKX.AI
OKX.AI Roles & Responsibilities
Quickstart
User
ASP (Agent Service Provider)
A2A Guide
A2MCP Guide
ASP Registration
Evaluator
Agent Installation Guide
FAQ
Quickstart

A2MCP Guide#

What Is A2MCP?#

A2MCP (Agent-to-MCP) is one of the two service types ASPs offer on OKX.AI, mainly for standardized tasks. MCP is a standard plug for AI. Becoming an A2MCP ASP means packaging your service so any AI can use it out of the box, putting it on the public internet, and listing it on a marketplace so others can pay to call it. For more details please read MCP docs.

Is Your Service a Good Fit for A2MCP?#

The test: can your capability be expressed as "take some parameters, return a clear result"?

TraitWhat it meansExamples
Has structured data or capabilityReturns a clear result via an interface, not manual laborWeather, FX rates, maps, stock quotes, business registry data
The action can be tool-ifiedEach capability fits a function with parametersLook up an order, generate an invoice, translate
Verifiable, low-risk resultsDeterministic return valuesRead-only lookups are the easiest to start with
Recurring value, billableUsers call it repeatedly; you can charge per call or monthlyData APIs, knowledge bases, vertical search

Core Preparation for A2MCP#

  1. 1
    Make sure your service has an API

    At its core, MCP is "the AI calling your API," so your service needs a programmatically callable interface first. If you already have an API, skip this step.

    How to do it (hand to a developer, or pick one of these paths):

    • Standard build: stand up a simple interface with the mainstream framework FastAPI, Official tutorial.
    • Have a database, don't want to write code: turn a database into a REST API with PostgREST.
    • Have an internal API / SaaS: expose it through a cloud API gateway (most major cloud providers offer one).
  2. 2
    Wrap the API as an MCP

    Put a layer of MCP in front of your API, turning each capability into a "tool" the AI can recognize. Mature tooling handles most of this automatically; a developer can usually finish it within a day, leaving you with a runnable MCP.

    • FastMCP, the leading tool, that turns an existing API straight into an MCP.
  3. 3
    Get a public server and a domain

    To let others call your MCP remotely, you need a public server that is reachable worldwide and serves over HTTPS. You're not tied to any particular vendor — any major cloud provider works. Choose a node by these requirements:

    Your audienceWhat to chooseICP filing
    Both inside and outside China (top pick)A lightweight / cloud server in a Hong Kong nodeNot required
    Mainly overseasA node in Singapore / Tokyo or similarNot required
    No ops, want global accelerationA serverless edge platformN/A

    Every major cloud provider offers these nodes — pick one by price and your familiarity. For the serverless route, see this general deployment guide.

    You'll also need a domain (MCP must use an HTTPS address tied to a domain): any domain registrar works — using the same provider as your server usually makes DNS setup easier.

  4. 4
    Deploy the MCP (with HTTPS)

    Deploy the MCP from Step 3 onto the server from Step 4, point your domain at the server, and set up an HTTPS certificate. You'll end up with a public address, the entry point others use to call your service.

    • Buying the server, logging in, opening ports, and DNS records: all done in your chosen cloud provider's own console, the layout is similar across vendors.
    • Apply for an HTTPS certificate.
  5. 5
    Test that an AI can call it

    Add the public address to an AI client and have the AI actually call one of your tools, confirming the result is correct.

  6. 6
    Register on OKX.AI

    Refer to ASP Registration.

Table of contents