Run Your First MCP#
Now that your Agentic Wallet is ready, let's connect your AI Agent to its first MCP service. Through MCP (Model Context Protocol), your Agent can directly invoke a wide range of services from premium merchants without any login, and settle payments using its on-chain wallet.
Prerequisites: Install an AI Agent#
This guide uses Claude Code as an example. Onchain OS supports all major Agents (Cursor, OpenClaw, etc.).
1. Install Onchain OS for your Agent#
If you haven't installed Onchain OS yet, tell your Agent:
Install Onchain OS for me and register this agent
npx skills add okx/onchainos-skillsThe Agent will complete the installation automatically. For more details, see GitHub.
2. Login to Your Agentic Wallet#
Before calling any MCP service, your Agent needs an active wallet session to sign requests and pay for service usage.
Tell your Agent:
Login to Agentic Wallet with my emailEnter your email and wait for the verification code:
<email>Enter the OTP code:
<otp-code>3. Fund Your Agentic Wallet#
MCP services may charge a fee paid in USDT. Before your first call, top up your wallet.
Tell your Agent:
Show me my wallet address and current balanceSend a small amount of USDT to the displayed X Layer address. Once the transfer confirms on-chain, your wallet is ready to use.
4. Call Your First MCP Service#
Now you can use any MCP Marketplace service. Copy the service prompt from the provider's page and paste it into your Agent. For example:
I’d like to use the service provided by Agent CoinAnk:
Service title: Get Trading Pair Real-Time Price
Service type: A2MCP
Endpoint: https://open-api.coinank.com/api/instruments/getLastPrice
Please send a request to this endpoint.Your Agent will:
- Resolve the service endpoint
- Sign the request with your Agentic Wallet
- Pay the service fee (if any)
- Return the response
5. Success#
Once the call completes, you'll see the service response inline in your Agent. From here, you can chain MCP services together — pulling signals, executing trades, pushing notifications — all from a single natural-language prompt.