Connectors

AgentOS provides four connectors that link AI agents to Airwallex APIs and documentation. Use this page to choose the right connector for your task and install it in your AI assistant.

Choose a connector

CLI versus MCP

The Airwallex CLI has the broadest API surface and is the right choice for power users and scripted automation in a terminal or CI environment. The MCP servers are required when the agent runs inside a chat UI or non-terminal context — for example, Claude Desktop, Cursor agent mode, Lovable, V0, or Repl.it — where the agent cannot shell out to a CLI.

For the bundled experience that includes both connectors plus pre-built workflows, install the AgentOS plugin.

Benefits

• Faster information retrieval: Agents fetch documentation and API references directly in the coding environment, with access to your codebase for context.
• Reduced hallucination: Connecting agents to current Airwallex documentation and API surfaces reduces the rate of incorrect or outdated code.
• Workflow automation: Common tasks — creating test resources, simulating events, executing production reads — run through natural language prompts without context-switching.
• Production-grade: Connectors are designed for use against live Airwallex accounts, not just demos. Money-out actions remain off by default; see Safety and responsibility.

Limitations

• Model dependency: Quality varies by model. Different AI assistants produce different results, and no model is perfect.
• Sandbox-only Developer MCP: Direct API testing tools in Developer MCP run in sandbox only. Production operations require the CLI or Production MCP (coming soon).
• Hallucination reduction, not elimination: Generated code and recommendations must always be reviewed, validated, and tested before use in production.

Install the airwallex CLI

The Airwallex CLI is a single self-contained binary for macOS and Linux. It auto-detects your OS and architecture and installs to ~/.local/bin/airwallex:

1curl -fsSL https://static.airwallex.com/developer-tools/airwallex-cli/install.sh | sh

After installation, sign in to your Airwallex account:

Set up

1# Login to Sandbox by default. Complete the authentication + authorization flow from the browser.
2airwallex auth login
3
4# Login to Production
5airwallex auth login --prod
6
7# Scope down if you want to control CLI's access
8# You can find the list of scopes here: https://www.airwallex.com/docs/developer-tools/partner-connections/oauth-scopes
9# Note that not all available scopes are required for the CLI 
10airwallex auth login --prod --scope=balance:read
11
12# Check environment and authentication status
13airwallex auth whoami
14
15# Clear authentication state
16airwallex auth logout
17
18# Send feedback from your device's default email client
19airwallex feedback --message "Can you please add support for listing <resource> from my organization?"

Usage

1# Let agent discover the entire CLI surface
2airwallex --tree
3
4# Let agent discover the sub-tree under a given CLI command
5airwallex --tree invoices
6
7# Review API request corresponding to a resource command before executing
8airwallex --dry-run products list
9
10# Compact JSON for efficient agent consumption
11airwallex --compact meters list --page-size=1
12
13# Write actions require confirmation in interactive mode by default.
14# Override when you know what you are doing
15echo '{
16  "name": "Demo product — ceramic coffee mug (12 oz)",
17  "description": "Example catalog item for testing the CLI. Not a real product or offer.",
18  "request_id": "ceramic-coffee-mug-245abe6"
19}' | airwallex --confirm products create --data-stdin
20
21# Discover the schema of the API endpoint invoked by a given resource command
22airwallex products update --api-schema-only

For the full command reference, run airwallex -h.

Telemetry

Telemetry is enabled by default. You can disable telemetry either globally using the DISABLE_TELEMETRY environment variable or for a specific command execution using the --no-telemetry global CLI option.

When enabled, the CLI collects basic device information (OS, architecture) & usage metrics related to command execution (command name, error code, latency, exit code) etc.

Install Developer MCP

Developer MCP is the sandbox MCP server. It exposes documentation tools, sandbox simulation tools, and certain endpoints.

Before you begin

• A compatible AI assistant installed (Cursor, Claude Code, or other MCP-compatible tool).
• Network access to https://mcp-demo.airwallex.com.
• A sandbox account if you want to use the remote endpoint or the local server.
• Sandbox API credentials (Client ID and API Key) if you want to use the local server.
• Node.js version 18.x or later installed and available on your PATH if you want to use the local server.

Cursor

You can use one-click installation links or manual configuration.

Remote
Local

Documentation and sandbox tools at the remote endpoint. Cursor initiates OAuth to authenticate with Airwallex.

Or manually add to your Cursor MCP settings:

JSON

1{
2  "mcpServers": {
3    "airwallex-dev": {
4      "type": "http",
5      "url": "https://mcp-demo.airwallex.com/developer"
6    }
7  }
8}

Full tool access including billing tools, without OAuth.

Or manually add the following to your Cursor MCP settings and fill in your sandbox credentials:

JSON

1{
2  "mcpServers": {
3    "airwallex-dev": {
4      "command": "npx",
5      "args": ["-y", "@airwallex/developer-mcp@latest"],
6      "env": {
7        "AIRWALLEX_SANDBOX_CLIENT_ID": "",
8        "AIRWALLEX_SANDBOX_API_KEY": "",
9        "AIRWALLEX_SANDBOX_ACCOUNT_ID": ""
10      }
11    }
12  }
13}

Claude Code

You can use one-click installation commands or manual configuration.

1claude mcp add-json airwallex-dev '{ "type": "http", "url": "https://mcp-demo.airwallex.com/developer" }'

1claude mcp add-json airwallex-dev '{ "type": "stdio", "command":"npx", "args": ["-y", "@airwallex/developer-mcp@latest"], "env": { "AIRWALLEX_SANDBOX_CLIENT_ID": "", "AIRWALLEX_SANDBOX_API_KEY": "", "AIRWALLEX_SANDBOX_ACCOUNT_ID": "" } }'

Claude.ai / Claude Desktop

Connect using the built-in Airwallex Developer connector.

Find the official Airwallex Developer connector from Settings > Connectors.

When you try to connect, Claude initiates an OAuth flow to authenticate with Airwallex.

Gemini CLI

1{
2  "mcpServers": {
3    "airwallex-dev": {
4      "type": "http",
5      "url": "https://mcp-demo.airwallex.com/developer"
6    }
7  }
8}

1{
2  "mcpServers": {
3    "airwallex-dev": {
4      "command": "npx",
5      "args": ["-y", "@airwallex/developer-mcp@latest"],
6      "timeout": 30000,
7      "env": {
8        "AIRWALLEX_SANDBOX_CLIENT_ID": "",
9        "AIRWALLEX_SANDBOX_API_KEY": "",
10        "AIRWALLEX_SANDBOX_ACCOUNT_ID": ""
11      }
12    }
13  }
14}

OpenAI Codex

1[mcp_servers.airwallex-dev]
2type = "http"
3url = "https://mcp-demo.airwallex.com/developer"

1[mcp_servers.airwallex-dev]
2command = "npx"
3args = ["-y", "@airwallex/developer-mcp@latest"]
4
5[mcp_servers.airwallex-dev.env]
6AIRWALLEX_SANDBOX_CLIENT_ID = ""
7AIRWALLEX_SANDBOX_API_KEY = ""
8AIRWALLEX_SANDBOX_ACCOUNT_ID = ""

Lovable

Lovable is a web-based coding agent that only supports remote MCP servers. MCP server support requires a paid plan. Connect using the MCP server settings in your Lovable project. Lovable initiates OAuth to authenticate with Airwallex.

V0

V0 by Vercel is a web-based AI coding agent. V0 does not support the Developer MCP remote endpoint today. Use the Docs MCP for documentation-only access, or another coding agent if you need sandbox tools.

Repl.it

Repl.it is a web-based AI coding agent that only supports remote MCP servers. Connect using the MCP server settings in your Repl.it project. Repl.it initiates OAuth to authenticate with Airwallex.

Install Docs MCP

Docs MCP is a documentation-only MCP server. It exposes the same documentation search tools as Developer MCP and requires no authentication. Use Docs MCP when your agent only needs to look up Airwallex docs, or when authentication is not available in the host environment.

Cursor

Or manually add to your Cursor MCP settings:

1{
2  "mcpServers": {
3    "airwallex-docs": {
4      "type": "http",
5      "url": "https://mcp-demo.airwallex.com/docs"
6    }
7  }
8}

Claude Code

1claude mcp add-json airwallex-docs '{ "type": "http", "url": "https://mcp-demo.airwallex.com/docs" }'

Claude.ai / Claude Desktop

Add a custom connector. From Settings > Connectors > Add custom connector, use https://mcp-demo.airwallex.com/docs as the Remote MCP server URL.

Gemini CLI

1{
2  "mcpServers": {
3    "airwallex-docs": {
4      "type": "http",
5      "url": "https://mcp-demo.airwallex.com/docs"
6    }
7  }
8}

OpenAI Codex

1[mcp_servers.airwallex-docs]
2type = "http"
3url = "https://mcp-demo.airwallex.com/docs"

Lovable

Configure the MCP server in your Lovable project settings using https://mcp-demo.airwallex.com/docs.

V0

1{
2  "mcpServers": {
3    "airwallex-docs": {
4      "type": "http",
5      "url": "https://mcp-demo.airwallex.com/docs"
6    }
7  }
8}

Repl.it

Configure the MCP server in your Repl.it project settings using https://mcp-demo.airwallex.com/docs.

Production MCP

Production MCP will let agents read and write production Airwallex resources from non-terminal environments where the Airwallex CLI cannot run — for example, hosted chat UIs and agent platforms. It will share the safety guardrails of other AgentOS components, including no money-out actions by default. Authentication will use OAuth against your production Airwallex account.

Production MCP is coming soon. In the meantime, use the Airwallex CLI for production reads and writes.

Configure the local server

For Developer MCP local mode, add the following environment variables to the env block in your MCP settings and fill in your sandbox credentials:

Verify your connection

After setup, confirm the connector is working:

1. CLI: run airwallex auth whoami to confirm the environment and whether you are signed in.
2. MCP: check that the Airwallex server appears in your editor's MCP configuration or settings panel.
3. Send a test prompt that explicitly mentions Airwallex, for example: "Using Airwallex, show me how to create a test payment link in the sandbox."
4. If the response references Airwallex documentation or invokes an Airwallex tool, the connection is working.

Next steps

• Build with AgentOS — prompting best practices, tools reference, and troubleshooting.
• Skills — pre-built workflows that use the connectors under the hood.
• Install plugins — get the CLI and skills into Claude Code or Cursor in one step.