AI agents with SePay

Use SePay from your AI agent. Look up transactions over MCP in ChatGPT and Claude, add SePay Skills to coding agents, and use SePay CLI to test webhooks and automate lookups.

||

AI agents such as Claude Code, Codex, Cursor and ChatGPT work with SePay through SePay MCP, SePay Skills and SePay CLI. They serve you while building and looking things up; your app still calls the SePay API and receives SePay Webhooks as before.

Choose SePay MCP, Skills or CLI

Pick by what you want the agent to do: look up data, write code or test your integration. You don't have to pick just one: use a single tool, combine two, or all three as you need, for example Skills so the agent writes the code and CLI to test webhooks on your machine.

Look up account data with SePay MCP

Say you want your monthly cash flow, or whether order DH00012 has been paid, without searching through my.sepay.vn.

You can use SePay MCP to ask Claude or ChatGPT in plain words and get answers from your own SePay account. While coding in Claude Code or Codex, the agent can look up data too, such as "Did any webhook fail this week?". When you ask how to integrate, the agent reads the SePay docs, which needs no sign-in. The agent picks the data it needs, within the access you granted.

The same connection works in chat apps and coding agents:

Show my monthly cash flow on SePay
Called get_cashflow_report

No activity from July onward. The peak month is May (15,983,000 VND in); the lowest is June (13,177,000 VND in).

Total incoming
89,652,000
Total outgoing
7,120,000
Net
82,532,000
IncomingOutgoing
16 M12 M8 M4 M0 M
JanFebMarAprMayJun

Connect Claude, ChatGPT, Claude Code or Codex to your SePay account.

Connect SePay MCP

Ask how to integrate SePay with SePay Skills

Say you want your website to confirm orders when customers pay by bank transfer, but you don't know where to start or how to do it right.

Add SePay Skills to ChatGPT or Claude, then ask the way you would ask a support agent: the AI tells you whether to install a plugin or integrate yourself, walks you through my.sepay.vn using the exact button names, and explains webhooks in plain words.

ClaudeChatGPTGeminiGrok…and coding agents such as Claude Code, Codex, CursorApps that take sepay.zip

Three steps: download the official SePay sepay.zip, click your profile picture, then Settings › Customize › Skills in your AI chat app, drag and drop sepay.zip into the upload box, the app turns the sepay skill on, then ask how to confirm store payments automatically.

To add the file to each app, see Install SePay Skills in AI chat apps.

Developers can install the skill in coding agents such as Claude Code, Codex, Cursor and other agents that support skills. When you ask for a SePay integration, the agent writes code that follows the SePay docs and the safety rules: sandbox by default, secret keys kept on the server, webhook signatures checked, each transaction processed only once.

Add the SePay skill to the coding agent you use.

Install for a coding agent

Test your integration and look up data with SePay CLI

Say you are integrating webhooks on your own computer, with the handler running on localhost:3000, and want to try it right away, with no server deploy and no real transfer.

You can use SePay CLI: terminal 1 receives webhooks on your machine, terminal 2 simulates a 50,000 VND incoming transaction with order code DH1709 (--bank-account takes the ID of your sandbox bank account; list them with sepay bank-account list). The webhook for the transaction reaches your code, and terminal 1 prints the HTTP status it returned:

Terminal 1: receive webhooks
$ sepay webhook listen --forward localhost:3000env: sandbox  Delivering for every bank account of this company.✓ Listening for sandbox events → http://localhost:3000/   (Ctrl+C to stop · retries+enter to manage)14:22:07  payment_in   → http://localhost:3000/   200   12ms   txn=a41c9e02-xxxx-xxxx-xxxx-fa163e2f7c98   wh=5f0e2a7b-xxxx-xxxx-xxxx-3c9d81e04b6a
Terminal 2: simulate a transaction
$ sepay sandbox simulate payment_in --bank-account 8f3c2a10-xxxx-xxxx-xxxx-0c4e7b9d2f61 --amount 50000 --code DH1709env: sandboxSimulated payment_in event.a41c9e02-xxxx-xxxx-xxxx-fa163e2f7c98

On production, the same commands look up transactions, create and update webhooks, set payment codes and update bank accounts without opening my.sepay.vn. For example, find the transactions with order code DH1709:

$ sepay transaction list --env production --search DH1709 --json[  {    "id": "07ad3dc9-xxxx-xxxx-xxxx-0afffe87053b",    "transaction_date": "2026-10-31 11:24:40",    "transfer_type": "in",    "amount_in": 50000,    "transaction_content": "DH1709 thanh toan don hang",    "code": "DH1709",    "bank_brand_name": "ACB",    ...  }]

Install the CLI and receive your first webhook on your machine.

Install SePay CLI

Limits and safety

SePay MCP: Read-only by default. Write actions run only when you grant access and confirm each one. The agent sees only the data you granted when connecting. Change or revoke access at any time on the AI Agent page (details).

With SePay CLI, a command that changes data on production runs only when you confirm it on your own terminal. An AI does not confirm it for you: it hands you the command to run yourself (confirmation rule).

Let agents read the SePay docs

An agent without SePay MCP, Skills or CLI can still read the SePay docs as plain text at these addresses:

SourceAddressUse when
Index for AI/llms.txtThe agent needs a list of every page with a short description
Full docs/llms-full.txtTools that load the whole doc set, for example to build an index. The file is very large (several MB), more than an agent session can hold, so agents should not read it directly
Markdown per pageAdd .md to any URL, for example /en/sepay-webhooks/tong-quan.mdThe agent needs one page only, faster and cheaper
SePay MCP docs toolslist_docs, fetch_contentThe AI app is connected to SePay MCP, no sign-in needed

Next