Airwallex AgentOS

Airwallex AgentOS is a toolkit for AI agents that need to act on your production Airwallex account — reading balances, creating beneficiaries, issuing cards, building cashflow reports, and running multi-step financial workflows on your behalf.

AgentOS bundles three things:

• Connectors — the Airwallex CLI and the Airwallex AgentOS MCP server, which connect agents to your production Airwallex account, plus the Airwallex Developer MCP for documentation lookups and sandbox testing.
• Skills — pre-built workflows for common tasks like contract-to-billing, beneficiary creation, card provisioning, and cashflow management.
• Plugins — one-step installation of the connectors and skills into Claude Code, Claude Cowork, and Cursor.

Install AgentOS

The fastest way to get AgentOS is to let your agent set it up for you. Paste the prompt below into any AI assistant — it will fetch this guide and handle the installation, so you don't have to run a series of commands yourself.

1Set up Airwallex AgentOS for me. Fetch https://www.airwallex.com/docs/developer-tools/ai/agentos.md and follow the setup guide.

Prefer to run the steps yourself? Each tab below is self-contained — copy the commands into your agent and it will install the connectors and skills for you.

1claude plugin install airwallex-agentos@claude-plugins-official

1curl -fsSL https://static.airwallex.com/developer-tools/airwallex-cli/install.sh | sh
2airwallex auth login --prod

1curl -fsSL https://static.airwallex.com/developer-tools/airwallex-cli/install.sh | sh
2airwallex auth login --prod

1codex plugin marketplace add https://github.com/airwallex/airwallex-marketplace
2codex plugin add airwallex-agentos@airwallex-marketplace

1curl -fsSL https://static.airwallex.com/developer-tools/airwallex-cli/install.sh | sh
2airwallex auth login --prod

1curl -fsSL https://static.airwallex.com/developer-tools/airwallex-cli/install.sh | sh
2airwallex auth login --prod

Which AgentOS connector do I need?

AgentOS provides three connectors. Pick based on what your agent needs to do and where it runs:

1Does your agent need to act on your production account?
2│
3├─ Yes, and it runs in a terminal (Claude Code, Cursor, Codex, CI, scripts)
4│   → use the Airwallex CLI
5│      Broadest API coverage. Best for power users, scripts, and CI.
6│
7├─ Yes, and it runs in a chat UI or hosted agent platform
8│   (Claude Cowork, Claude.ai, agent frameworks)
9│   → use the Airwallex AgentOS MCP
10│      Production reads and writes where the agent can't shell out to a CLI.
11│
12└─ No — it only needs docs lookups or sandbox testing
13    → use the Airwallex Developer MCP
14       Documentation and sandbox tools, no production access.

Airwallex AgentOS MCP

Airwallex AgentOS MCP lets 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 shares the safety guardrails of other AgentOS components, including no money-out actions by default. Authentication uses OAuth against your production Airwallex account.

Claude Code

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

Claude.ai / Claude Cowork

Find the official airwallex-agentos connector from Settings > Connectors and connect. Claude initiates an OAuth flow to authenticate with your production Airwallex account.

Cursor

You can use a one-click installation link or manual configuration. Cursor initiates OAuth to authenticate with Airwallex:

Or manually add to your Cursor MCP settings:

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

Other agents

Any MCP-compatible agent can connect to the remote endpoint https://mcp.airwallex.com/mcp and complete the OAuth flow against your production account.

Skills

AgentOS skills are pre-built workflows that teach AI agents how to complete common Airwallex tasks end-to-end. Each skill bundles the prompts, command sequences, and validation logic an agent needs to handle one workflow correctly using the Airwallex CLI. Skills are distributed through the Airwallex marketplace and installed alongside the connectors when you install AgentOS.

Available skills

After installation, restart your agent (or start a new session) and the skills are available. You can confirm by asking the agent to list its available skills, or by invoking one directly (for example, "use the beneficiary-creation skill to add this supplier").

Recommended model

Skills coordinate multi-step workflows that involve document extraction and API orchestration. For best results, run them with Claude Opus 4.6 or later.

Airwallex Developer MCP

Airwallex Developer MCP gives agents documentation lookups and sandbox testing tools, so you can build and validate integrations alongside the production workflows that the Airwallex CLI and Airwallex AgentOS MCP handle. It runs against the sandbox only and never touches your production account.

Connect any MCP-compatible agent to the remote endpoint https://mcp-demo.airwallex.com/developer and complete the OAuth flow against your sandbox account. For one-click install links, local-server setup, and the full tool list, see the developer connectors guide.

Who it's for

• Operators running finance workflows. Use the plugin in Claude Code, Claude Cowork, or Cursor to have an agent read your account, draft beneficiaries, provision cards, and build reports — with money-out actions off by default.
• Developers building custom agents. Use the Airwallex CLI or Airwallex AgentOS MCP directly from your own agent framework, and add the Airwallex Developer MCP for docs lookups and sandbox testing while you build.

Build with AgentOS

For prompting guidance, connector selection, and troubleshooting tips, see Build with AI.

Safety and responsibility

AgentOS connects AI agents to live Airwallex APIs. Before you deploy:

• Actions taken by agents using your Airwallex credentials are governed by the Airwallex Terms of Service .
• AgentOS components do not initiate money-out actions (transfers, FX conversions, or payouts) on your behalf. Where the underlying API supports them, the components are explicitly instructed not to execute them.
• Agent outputs are probabilistic. Always review generated code, configurations, and recommendations before acting on them in production.

Feedback

For questions or feedback about AgentOS, contact the Developer Experience team at [email protected].